Accepted papers

JURSE 2019 Accepted Papers

(99 accepted, 121 submissions, acceptance rate: 81%)

* Jakob Hartig, John Friesen and Peter F. Pelz. Spatial relations of slums: size of slum clusters

Abstract

More than half of the world’s population currently lives in cities. In many parts of the world, slums are part of the urban landscape. Despite different history, cultures and continents, slums share common properties. In this paper, the spatial relations of slums in Dhaka (Bangladesh) and Rio de Janeiro (Brasil) are analysed on different scales with the Average- Nearest-Neighbour and the Hopkins index. It is found that the size of the sample windows for the analysis has a huge influence on the form of the spatial relationship. On the city scale slums are clustered, while on smaller scales there is a relatively large variation in spatial relationships with predominantly random distribution. It is observed that there is a scale where the large variations of the spatial relationships vanish in a sharp transition. This scale is interpreted as a characteristic size of slum clusters, which has implications for understanding the emergence of slums and the optimum design of infrastructure for slums.

* Sergiy Kostrikov, Rostyslav Pudlo, Dmitry Bubnov and Anna Kostrikova. Studying of urban features by the multifunctional approach to LiDAR data processing
Abstract

Our paper focuses on the original multifunctional approach to LiDAR data processing. This approach implies further joint and composite implementation in urban studies all of obtained results that formally belong to three various modeling domains: Building Extraction (BE), Change Detection (CD), and Digital Elevation Model Generation (DEM-G). Our both feature extraction algorithms of the High Polyhedral Modeling (HPM) based on Delaunay refinement as “a covering TIN”, and the Low Polyhedral Modeling (LPM) based on the updated point segmentation and clustering are briefly introduced. HPM technique is intended to process aerial data as well as mobile ones. Both a web-based application, and a cloud-based one, as well as desktop software have been elaborated with this multifunctional processing methodology implementation. Few interface illustrations of these products, provided on open-source data, are referred. This study outlook supposes to extent the processing pipeline over an urban block, and a district scopes as far as to the whole city modeling space.

* Oussama Ennafii, Clément Mallet, Arnaud Le Bris and Florent Lafarge. The necessary yet complex evaluation of 3D city models: a semantic approach
Abstract

The automatic modeling of urban scenes in 3D from geospatial data has been studied for more than thirty years. However, the output models still have to undergo a tedious task of correction at city scale. In this work, we propose an approach for automatically evaluating the quality of 3D building models. A taxonomy of potential errors is first proposed. Handcrafted features are computed, based on the geometric properties of buildings and, when available, Very High Resolution images and depth data. They are fed into a Random Forest classifier for the prediction of the quality of the models. We tested our framework on three distinct urban areas in France. We can satisfactorily detect, on average 96% of the most frequent errors.

* Lijun Zhao and Ping Tang. Improved Visual Vocabularies for Scene Classification of High Resolution Remote Sensing Imagery in Urban Areas
Abstract

The improvement of spatial resolution of remote sensing images provides more and more ground details, which makes it possible to better interpret unban land use and analyze urban functional areas. The bag-of-visual-words (BOVW) model becomes a well-known method to deal with such land-use scene classification problems. Generally, the unsupervised k-means clustering method is used to construct visual dictionaries, in which there often exist large amounts of local features to cluster in the visual vocabulary construction stage, largely affecting the computational complexity and hindering the schedule of visual dictionary generation. To solve the above mentioned problems, this paper proposes a two-phase k-means clustering based visual vocabulary construction method. The proposed method can add information of predefined scene categories from different images, which is beneficial for the selection of initial cluster centers, and, to a certain extent, alleviates the problem that the amount of samples to be clustered in a single course of clustering can be too large. Experimental results on the public UCM-21 dataset show that compared with the traditional method, the proposed one can not only reduce the dictionary construction time but also help improve the classification accuracy.

* Uta Heiden, Andrea Marinoni and Paolo Gamba. Human settlement and infrastructure monitoring with hyperspectral imaging
Abstract

This paper reviews in detail the contributions of hyperspectral imaging to the topic of urban remote sensing. Hyperspectral imaging is traditionally connected to the spectral characterization of surface materials. Moreover, urban areas are characterized by a very complex geometrical structure, which requires either very high spatial resolution or complex unmixing procedures based on linear and non-linear mixing models. Non-linear unmixing and material mapping using both spectral and spatial features are therefore two important topics when using hyperspectral imaging to monitor human settlements and infrastructures. Finally, even when no specific material and or urban element is sought, the mixture of artificial (as opposed to natural) materials in human settlements can be used to delineate their extents, with excellent results with respect to those obtained by multispectral optical sensors with the same spatial resolution.

* Borja Rodríguez-Cuenca, Jesús María Garrido Sáenz de Tejada, Sara Lorite Martínez and Jesús Moreno Jabato. Cartography production and cadastral database updating using multi-temporal aerial laser scanner point clouds
Abstract

Automatic updating of cartographic databases is one of the main goals of cartographers since the origins of this scientific field. Detecting changes in land covers through visual inspection of the surface is costly in economic and temporal terms. Multitemporal remote sensing data ease and speed up the detection of changes in Earth surface. In this work it is proposed a method to identify changes in natural and anthropogenic land covers by comparing aerial laser scanner 3D point clouds registered in different dates, one in 2010 and other in 2016. The effectiveness of the proposed method was tested in the Spanish cities of Madrid and Barcelona. Results obtained in both test sites provide the new constructions and those buildings that have been demolished in the 6 years elapsed between both considered flights. Besides, it is proposed a new raster product created from LIDAR data that can be used as a base map and as a source to consult the heights of buildings and vegetation.

* Wei Zhang, Ping Tang and Lijun Zhao. A Comparative Study of U-Nets with Various Convolution Components for Building Extraction
Abstract

Building extraction from remote sensing imagery is always a nontrivial topic in the field of earth observation. Recently, approaches based on fully convolutional networks have achieved state-of-the-art performance in the semantic segmentation and made it possible to perform dense pixel-wise classification, among which the U-Net architecture is one of the most popular one. In this paper, three U-Nets consisting of the normal convolution block, the residual unit and the inception module, named Normal-U-Net, Residual-U-Net and Inception-U-Net respectively, are designed for building extraction. Experiments on the public Massachusetts building dataset show that the Inception-U-Net is the most competitive architecture comprehensively. In addition, the three single U-Nets are combined into a stronger model called EU-Net, which has achieved a remarkable performance for building extraction.

* Qinghui Liu, Arnt-Børre Salberg, Robert Jenssen and Michael Kampffmeyer. DENSE DILATED CONVOLUTIONS MERGING NETWORK FOR SEMANTIC MAPPING OF REMOTE SENSING IMAGES
Abstract

We propose a network for semantic mapping called Dense Dilated Convolutions Merging Network (DDCM-Net) to provide a deep learning approach that can recognize multi-scale and complex shaped objects with similar color and textures, such as buildings, surfaces/roads, and trees in very high-resolution remote sensing images. The proposed DDCM-Net consists of dense dilated convolutions merged with varying dilation rates. This can effectively enlarge the kernels’ receptive fields, and more importantly obtain rotational invariant information to promote surrounding discriminative capability. We demonstrate the effectiveness of the proposed DDCM-Net on the public ISPRS Potsdam labeling dataset and achieve a competitive performance of 92.8% overall accuracy, 92.2% F1-score and 85.8% mean intersection over union by only using the RGB bands, without any post-processing. We also show results on the ISPRS Vaihingen labeling dataset, where the DDCM-Net trained with IRRG bands, also obtained better mapping accuracy than previous state-of-the-art approaches.

* Yuansheng Hua, Lichao Mou and Xiao Xiang Zhu. Multi-label Aerial Image Classification using A Bidirectional Class-wise Attention Network
Abstract

Multi-label aerial image classification is of great significance in remote sensing community, and many researches have been conducted over the past few years. However, one common limitation shared by existing methods is that the co-occurrence relationship of various classes, so called class dependency, is underexplored and leads to an inconsiderate decision. In this paper, we propose a novel end-to-end network, namely class-wise attention-based convolutional and bidirectional LSTM network (CA-Conv-BiLSTM), for this task. The proposed network consists of three indispensable components: 1) a feature extraction module, 2) a class attention learning layer, and 3) a bidirectional LSTM-based sub-network. Experimental results on UCM multi-label dataset and DFC15 multi-label dataset validate the effectiveness of our model quantitatively and qualitatively.

* Sylvain Lobry and Devis Tuia. Deep Learning Models to Count Buildings in High-Resolution Overhead Images
Abstract

This paper addresses the problem of counting buildings in very high-resolution overhead true color imagery. We study and discuss the relevance of deep-learning based methods to this task. Two architectures and two loss functions are proposed and compared. We show that a model enforcing equivariance to rotations is beneficial for the task of counting in remotely sensed images. We also highlight the importance of robustness to outliers of the loss function when considering remote sensing applications.

* Eike Jens Hoffmann, Martin Werner and Xiao Xiang Zhu. Building Instance Classification using Social Media Images
Abstract

Understanding urbanization and planning for the upcoming changes require detailed knowledge about the places where people live and work. Thus, knowing the usage of buildings is inevitable to distinguish between residential and commercial places. Assessing the usage of buildings from an aerial perspective alone is challenging and results in unresolveable ambiguities. As complementary data sources, social media images taken from ground level allow access to the building façades, as well as ongoing social activities around the buildings, which are very valuable information while coming to accessing the building usages. Towards the fusion of social media images and remote sensing data for this purpose, in this work we present a method to assess building usages from social media images taken in their neighborhood. Using a straight forward next neighbor classifier for mapping images to buildings and pre-trained networks for dimensionality reduction we trained a logistic regression classifier to distinguish between five different building usage classes. Applied to a study area of Los Angeles metropolitan area, USA, we obtain an average precision of 0.67. Hence, we show that social media images can be a valuable additional source to remote sensing data.

* Iphigenia Keramitsoglou, Klea Katsouyanni, Panagiotis Sismanidis, Angeliki Efstathiou, Eleni Myrivili, Nikos Bogonikolos and Chris T. Kiranoudis. Satellite-based Emergency Notification System to Support Cities During Extreme Temperature Events
Abstract

Remote sensing of urban areas can provide valuable data about the function of cities. The considerable technological improvements of the last decade provide new opportunities to use such data and open the way for the development of new applications and services. To that end, this work discusses the use of satellite thermal data in heat-health applications for disaster risk reduction (DRR). In particular, this paper presents how EXTREMA (EXTReme tEMperature Alerts for Europe), a 2018 funded European civil protection program, converts geostationary thermal image data into actionable information. This information has the form of personalized heatwave risk estimations that can be readily used by the general public and the city authorities.

* Yao Shen, Huanfeng Shen, Qing Cheng, Liwen Huang and Liangpei Zhang. Urban Expansion Trajectories in China’s 36 Major Cities
Abstract

As the largest developing country, China has experienced dramatic urban sprawl since the “reform and opening-up” policy started at the end of the 1970s. To find out the laws of the past urbanization in China is of great importance for promoting a sustainable development in the future. In this paper, we monitor three decades of urban expansion in China’s 36 major cities, based on the spectral mixture analysis of remotely sensed satellite images. The results demonstrate that these major cities have expanded by 5.85 times from 1986 to 2015, with 15.51km2 average expansion area per city per year. We find the urban expansion trajectories show three different modes, i.e., exponential, linear and s-shaped, which are closely related to the city development level. In addition, there is an interesting common tendency of the impervious surface first increasing and then decreasing in the old city zones.

* Ahmed Nassar, Sébastien Lefèvre, Jan Wegner and Nico Lang. Learning geometric soft constraints for multi-view instance matching across street-level panoramas
Abstract

We present a new approach for matching tree instances across multiple street-view panorama images for the ultimate goal of city-scale street-tree mapping with high positioning accuracy. What makes this task challenging is the strong change in viewpoint, different lighting conditions, high similarity of neighboring trees, and variability in scale.

We propose to turn (tree) instance matching into a learning task, where image-appearance and geometric relationships between views fruitfully interact. Our approach constructs a siamese convolutional neural network that learns to match two views of the same tree given many candidate tree image cut-outs and geographic information of the two panorama images. In addition to image features, we propose utilizing location information about the camera and the tree. Our method is compared to existing patch matching methods to prove its edge over state-of-the-art. This takes us one step closer to the ultimate goal of city-wide tree mapping based solely on panorama imagery to benefit city administration.

* Ahmed Nassar and Sébastien Lefèvre. Automated Mapping Of Accessibility Signs With Deep Learning From Ground-level Imagery and Open Data
Abstract

In some areas or regions, accessible parking spots are not geolocalized and therefore both difficult to find online and excluded from open data sources. In this paper, we aim at detecting accessible parking signs from street view panoramas and geolocalize them. Object detection is an open challenge in computer vision, and numerous methods exist whether based on handcrafted features or deep learning. Our method consists of processing Google Street View images of French cities in order to geolocalize the accessible parking signs on posts and on the ground where the parking spot is not available on GIS systems. To accomplish this, we rely on the deep learning object detection method called Faster R-CNN with Region Proposal Networks which has proven excellent performance in object detection benchmarks. This helps to map accurate locations of where the parking areas do exist, which can be used to build services or update online mapping services such as Open Street Map. We provide some preliminary results which show the feasibility and relevance of our approach.

* Dino Ienco, Kenji Ose and Christiane Weber. Towards combining Satellite Imagery and VGI for Urban LULC classification
Abstract

In this work we introduce and evaluate a deep learn- ing model, multi-branch CNN (mbCNN), that combines together satellite imagery and Volunteer Geographical Information (VGI) data to deal with Land Use/Land Cover (LULC) classes involving different types of built-up surfaces. Differently from most of the previous works that only consider Urban/Non-Urban settings or scenarios involving only one urban LULC class, here, we investigate the possibility to go a step further and distinguish among several urban land use classes: residential, industrial, sport fields and non-urban.

Experiments on a real-world dataset covering the City of Montpellier (South of France) site are reported. Such results demonstrate the quality of Deep Learning approaches to deal with several types of Urban LULC mapping as well as the positive influence to integrate VGI knowledge in the process.

* Jeremiah J. Nieves, Jessica E. Steele, Alessandra Carioli, Maksym Bondarenko, Alessandro Sorichetta, Andrew J. Tatem, Catherine Linard, Forrest R. Stevens and Andrea E. Gaughan. Annual Mapping of Human Settlements in the Absence of Remotely Sensed Urban Features
Abstract

Regular, e.g. annual, settlement extents datasets have a wide variety of applications in demography, public health, sustainable development, for example. Recently more multitemporal urban feature or human settlement datasets have become available, however, gaps in remotely sensed imagery coverage, cloud cover, and or expenses involved in producing such feature sets prevent more frequent temporal coverage while maintaining high spatial resolution. Here we demonstrate an interpolative and flexible modeling framework for producing annual built-settlement extents using a combination of random forest and spatio-temporal dasymetric techniques with subnational data to produce annual 100m x 100m resolution binary settlement maps in Uganda. We find that random forests are a strong technique for predicting transitions from non-settlement to settlement (1.53% out-of-bag error rate) and that when compared to observed extents at least half of the subnational units performed better than a null model as measured by the F1 score.

* Javiera Castillo-Navarro, Nicolas Audebert, Alexandre Boulch, Bertrand Le Saux and Sébastien Lefèvre. What Data are needed for Deep Learning in Earth Observation?
Abstract

This paper explores different aspects of semantic segmentation of remote sensing data using deep neural networks. Learning with deep neural networks was revolutionized by the creation of ImageNet, a massive, annotated image dataset. Remote sensing benefited of these new techniques, however Earth Observation datasets remain small in comparison. In this work we investigate how we can progress towards the ImageNet of remote sensing. In particular, two main questions are addressed in this project. First, how robust are existing supervised learning strategies with respect to data? Second, is it possible and what would be the benefits to create a dataset equivalent to ImageNet for remote sensing data? The main contributions of this work are: a strong analysis of the robustness of existing supervised learning strategies with respect to remote sensing data, the introduction of a new dataset, the MiniFrance dataset, as a first step towards the ImageNet of remote sensing.

* Caroline M. Gevaert, Divyani Kohli and Monika Kuffer. Challenges of mapping the missing spaces
Abstract

Urbanization in many global-south regions is often characterized by the proliferation of deprived neighborhoods (often referred to as slums). The importance of improving the lives of the residents in these areas is highlighted by many global development agendas. Unfortunately, improvement efforts are hampered by a lacking, inaccessible, or outdated spatial data. In this paper, we describe the current limitations which should be addressed to enable a widespread scaling up of remote sensing and image processing methodologies capable of providing this data. We focus on the conceptual ambiguity of what is understood as a slum, informal settlement, or deprived neighborhood. There is a wide diversity of their appearance within a single city, not to mention at a global scale. This leads to existential and extensional uncertainty, causing even experts to have different views of a slum’s boundaries when presented with the same image. Such conceptual ambiguities make it more difficult to obtain training data for image processing algorithms, as well as validation to test their accuracy. This also makes it difficult to improve the geographic, contextual, and temporal transferability of the algorithms. After discussing what is needed to upscale current algorithms, we continue to describe the gap between the geospatial data products the remote sensing community develops and the information needed by policymakers and other user-groups. We discuss why an objective and transparent system for monitoring slums is needed to monitor global development goals as well as support local communities and NGOs.

* Marie-Leen Verdonck, Frieke Van Coillie, Hans Hooyberghs, Frederik Priem and Matthias Demuzere. Spatial characterisation of heat risk in the Brussels Capital Region, Belgium
Abstract

Urban residents are exposed to higher levels of heat stress in comparison to the rural population. For the city of Brussels, we explore the influence of urban planning and global greenhouse gas emissions (GHG) for the near (2031-2050) and far (2081-2100) future. We implemented two urban planning expansion scenarios (translated into Local Climate Zones, LCZ) and two Representative Concentration Pathways (RCPs 4.5 and 8.5). The projections show that the influence of GHG emissions trumps urban planning measures in each of the two periods. In the near future, no large differences are noted between the RCP scenarios. In the far future on the contrary, both heat stress and risk values are twice as large for RCP 8.5 compared to RCP 4.5. Depending on the GHG scenario and the LCZ, heat stress is projected to increase with a factor of 10 by 2090 compared to the present-day climate and urban planning conditions. The imprint of vulnerability and exposure is clearly visible in the heat risk assessment, leading to very high levels of heat risk most notable for the north-western part of the Brussels Capital Region (BCR). The results demonstrate the need for mitigation and adaptation plans at different policy levels that strive for lower GHG emissions and the development of sustainable urban areas safeguarding livability in future cities.

* Thomas Stark, Michael Wurm, Hannes Taubenböck and Xiao Xiang Zhu. Slum Mapping in Imbalanced Remote Sensing Datasets Using Transfer Learned Deep Features
Abstract

Unprecedented urbanization, particularly in countries of the Global South, results in the formation of slums. Here remote sensing has proven to be an extremely valuable and effective tool for mapping slums. Recent advances in transferring deep features learned in fully convolutional networks (FCN) allow for mapping the specific structural types and alignments of buildings in slums. Thus, in our study we aim to analyze the transfer learning capabilities of FCNs for slum mapping with respect to training on imbalanced datasets and the quantity of available training images. When increasing the class imbalance of slums an improvement of the Intersection over Union (IU) of $\mathbf{10\%\textbf{ to }30\%}$ is possible. Increasing the total number of images improves the IU up to $\mathbf{20\%\textbf{ to }50\%}$. Transfer learning is especially useful when training on a combined dataset resulting in an IU of $\mathbf{71\%}$ with only a tenth of the original dataset available for training. Thus, transfer learning proofs extremely valuable in retrieving information on complex and heterogeneous urban structures such as slum patches.

* Henri Debray, Monika Kuffer, Claudio Persello, Christien Klaufus and Karin Pfeffer. Detection of Informal Graveyards in Lima using Fully Convolutional Network with VHR Images
Abstract

Lima is facing rapid urban growth, including a rapid expansion of informal areas, mainly taking place within three peripheral cones. Most of the studies on that subject focused in general on informal settlements. Yet in this paper, we focus on two different types of informal developments, graveyards and housing. They are experiencing complex, intertwined development dynamics due to a lack of land for housing and burials, causing social and public health problems. Land invasions on burials grounds have never been systematically investigated. Yet, while challenging due to their morphological similarity, the detection of boundaries between graveyards and informal housing is essential, e.g., to prevent the spread of diseases. This study aims to distinguish those similar urban structures of which the visual features are very alike (e.g., rectangular shapes, same colours, organic organization). We used state-of-the-art Fully Convolutional Networks (FCNs) with dilated convolution of increasing spatial kernels to acquire features of deep level of abstraction on Pleiades satellites. We found that such neural networks can reach a good level in mapping both informal developments with a F1-score of 0.819. Effective monitoring of such developments is important to inform planning and decision-making processes to allow interventions at critical locations.

* Sarochinee Kaewthani and Chaiyapon Keeratikasikorn. Improving the SLEUTH urban growth model via temporal consistency in urban input data
Abstract

Changes in an urban growth model were investigated after processing temporal consistency evaluation of classified urban images. A consistency evaluation involving both temporal filtering and heuristic reasoning was then applied to sequence classification of urban maps for further improvement. The SLEUTH urban growth model was tested in regions of uncontrolled urban expansion. The SLEUTH was calibrated using data collected from the major urban area of Nakhon Ratchasima, Thailand in 1989, 1994, 1999 and 2005. The best value of Optimal SLEUTH Metric (OSM) was calculated for urban inputs with and without temporal consistency checking. OSM value higher than without, presenting a better explanation of urban growth in the study area.

* Panagiotis Sismanidis, Iphigenia Keramitsoglou, Anastasia Tsontzou, Benjamin Bechtel, Stefano Barberis and Chris T. Kiranoudis. A Satellite-derived Heating- and Cooling-Degrees Geospatial Dataset: Results for Antwerp
Abstract

In the context of developing a low-carbon economy by 2050, the European Union (EU) member states have committed to improve energy efficiency by at least 27% by 2030. To empower municipalities and local authorities addressing this goal, the H2020 PLANHEAT research project develops an open-source software tool for developing economically sustainable energy plans for low-carbon heating and cooling. To take into account the Urban Heat Island (UHI) effect on energy demand, the PLANHEAT software tool uses a geospatial dataset of hourly 1 km Heating and Cooling Degrees that is derived from satellite thermal data and information from weather models. This article describes the methodology used for producing this dataset and presents the first results for the city of Antwerp in Belgium. The PLANHEAT tool will be released as a QGIS plugin in June 2019.

* Deepank Verma, Arnab Jana and Krithi Ramamritham. Identifying degrees of built arrangement in Indian cities through mid-resolution satellite imagery and Convolutional Neural Networks
Abstract

In recent years, the capability of Convolutional Neural Networks (CNN) in solving computer vision problems has been experimented widely. However, such studies have not been translated to the large-scale remote sensing approaches. The performance of CNN in satellite image classification tasks has been found superior to that of traditional algorithms. However, comparatively fewer studies have considered open earth observation mid and low-resolution datasets to classify intra-urban characteristics. This study utilizes the CNN model to identify the degrees of built arrangement in mid resolution Sentinel 2B imagery downloaded for ten largest Indian cities. Training and testing datasets for seven land cover classes including compact, open and sparse built form are created with the help of Google Earth platform. The definitions of the classes are taken from the LCZ classification scheme. The classification results are plotted for each city and compared with each other.

* Dorothee Stiller, Thomas Stark, Michael Wurm, Stefan Dech and Hannes Taubenböck. Large-scale building extraction in very high resolution aerial imagery using Mask R-CNN
Abstract

Urban areas are hotspots of complex and dynamic alterations of the Earth’s surface. Using deep learning techniques in remote sensing applications can significantly contribute to document these tremendous changes. Open source building data at a very high level of detail are still scarce or incomplete for many regions, therefore, hindering research and policy to properly provide knowledge on urban structures. In this study, we use a convolutional neural network to extract buildings for the city of Santiago de Chile. We deploy the recently released Mask R-CNN and use a pretrained model (PM) which already has been trained with remote sensing imagery. We fine-tune PM with very high resolution (VHR) airborne RGB images from our study region and generate the fine tuned model (FM). To extend the number of training data we test several data augmentation methods for training purposes and evaluate their performance. We achieve highest overall accuracy of 92% by using augmentations and the generated FM. The results underpin the high relevance of data augmentation techniques to increase training data size, and, thus, achieve higher classification accuracy. Our presented method can be adapted and applied to other global urban regions, and, help to overcome lacks in open source building data to assess urban environments.

* Nicolas Johannes Kraff, Hannes Taubenböck and Michael Wurm. How dynamic are slums? EO-based assessment of Kibera’s morphologic transformation
Abstract

Urban morphologies change over time. The dynamics and nature of morphological changes in informal settlements or slums have largely not been scientifically investigated. Consequently, it is necessary to fill the gap of the international demand for timeline analysis. In this work, earth observation (EO) is used to discover morphologic changes within eight years (2006-2014) in Nairobi’s major slum Kibera. Research mostly handles automated detection but in this study the classical visual image interpretation is applied on a very high level of detail capturing buildings in three dimensions. Consistencies and deviations in time are measured according to morphological variables. We find dynamics in the slum area high in terms of a 77% rise in number of buildings; at the same time, density increases only by 10%. Overall, the general pattern of complex, organic structure remains mostly unchanged.

* Elias Mendez Dominguez, David Small and Daniel Henke. Synthetic Aperture Radar Tomography for Change Detection Applications
Abstract

Tomographic synthetic aperture radar (TomoSAR) can broaden the scope of change detection applications for urban studies, human activity and forest monitoring. In this work we design and evaluate a method utilizing SAR tomography for change detection purposes applied to human activity monitoring and urban studies. The method uses 2-D images to detect changes caused by targets with a small vertical extent, and 3-D images for changes caused by targets with a large vertical extent. It exploits both amplitude and height difference information combined in a conditional random field to detect changes of interest. A significant performance improvement was obtained when comparing to methods using 2-D or 3-D images only.

* Yilei Shi, Qingyu Li and Xiao Xiang Zhu. BFGAN — building footprint extraction from satellite images
Abstract

Building footprint information is an essential ingredient for 3-D reconstruction of urban models. The automatic generation of building footprints from satellite images presents a considerable challenge due to the complexity of building shapes. In this work, we have proposed improved generative adversarial networks (GANs) for the automatic generation of building footprints from satellite images. We used a conditional GAN with a cost function derived from the Wasserstein distance and added a gradient penalty term. The achieved results indicated that the proposed method can significantly improve the quality of building footprint generation compared to conditional generative adversarial networks, the U-Net, and other networks. In addition, our method nearly removes all hyperparameter tuning.

* Yilei Shi, Yuanyuan Wang, Xiao Xiang Zhu and Richard Bamler. Large-Scale Urban Mapping using Small Stack Multi-baseline TanDEM-X Interferograms
Abstract

Multi-baseline synthetic aperture radar (SAR) interferometric techniques, such as SAR tomography, is well established for 3-D reconstruction in the urban area. These methods usually require fairly large interferometric stacks (> 20 images) for a reliable reconstruction. They are not directly applicable to SAR interferometric (InSAR) stack with only a few acquisitions, as the extremely small number of acquisitions can severely bias the estimates from the spectral estimators, such as beamforming which is often only asymptotically optimal. In addition, the number of images also causes severe ambiguity issue of the pixel with low signal-to-noise ratio. In this work, we propose a new processing framework of 3-D reconstruction with TomoSAR using extremely small stacks. Moreover, the applicability of the algorithm is demonstrated by exploiting TanDEM-X co-registered phase preserving single look slant range complex SAR images (CoSSC) over a large-scale test site of the whole Munich city, Germany. The reconstructed results have been systematically compared with global production of TanDEM-X digital elevation models (DEM) and LiDAR dataset, which show the potential of high quality large-scale 3-D urban mapping.

* Hannes Taubenböck, Felix Dahle, Christian Geiß and Michael Wurm. Europe’s socio-economic disparities reflected in settlement patterns derived from satellite data
Abstract

Development across Europe is uneven. This inequality finds its expression e.g. in different settlement pattern characteristics. We assume that increased settlement concentration reflects economic benefits. For capturing these patterns across the continent, we analyze a binary settlement map derived from Earth observation data. From it, we detect urban nodes as anchor points of urban densification. We identify a network of cities when conjugation lines between these urban nodes feature high settlement density. Further, we map regions around connected urban nodes with high settlement densities. We assume that these identified regions belonging to a network of cities express beneficial economic development. We test this hypothesis by assigning the economic indicators ‘gross domestic product’, ‘unemployment rate’, and ‘household income per inhabitant’ from Eurostat data sets to the identified regions belonging to a city network. We find economic advantages of these city network regions over other European regions.

* Stavros Stagakis, Ingunn Burud, Thomas Thiis, Niki Gaitani, Emmanouil Panagiotakis, Giannis Lantzanakis, Nektarios Spyridakis and Nektarios Chrysoulakis. Spatiotemporal monitoring of surface temperature in an urban area using UAV imaging and tower-mounted radiometer measurements
Abstract

An extended UAV campaign that covered the historic city center of Heraklion, Greece has been performed during two consecutive days in July 2018. Heraklion city center is equipped with a permanent micrometeorological tower measuring net radiation and turbulent heat and CO2 fluxes. UAV cameras with RGB/NIR bands, as well as an integrated thermal (IR) band were used during the campaign. A calibrated and orthorectified RGB/NIR image mosaic (5 cm) was produced and a Support Vector Machines (SVM) algorithm was applied for the classification of the urban surface materials. The emissivity of the different materials was obtained by spectral libraries and used to calibrate the thermal maps (10 cm) by the IR camera. The first results show the pronounced effects of urban canyon orientation, building density and materials to the surface temperature. Cool materials present significantly lower temperature than other materials. The surface temperature maps by the UAV were evaluated using tower-mounted net radiometer measurements. Increased differences between the two methods were found, attributed mainly to the different field of view of the two instruments, the increased thermal anisotropy of the urban environment and the uncertainty regarding the emissivity of the different materials.

* Josselin Aval, Jean Demuynck, Emmanuel Zenou, Sophie Fabre, David Sheeren, Mathieu Fauvel and Xavier Briottet. Identification of the London plane in urban alignment based on hyperspectral data and contextual information
Abstract

This paper presents an approach for identifying the London plane in urban alignment based on hyperspectral data and contextual information. The proposed approach allows the London plane street trees of a study case to be perfectly identified thanks to both a supervised classification and a post regularization of the species prediction, based on the alignment membership derived from a Marked Point Process approach.

* Andreas Braun, Gebhard Warth, Felix Bachofer and Volker Hochschild. Identification of roof materials in high-resolution multispectral images for urban planning and monitoring
Abstract

Monitoring and characterizing cities is a key challenge of earth observation. While the application of very high-resolution imagery is widely acknowledged, the use of sensors of spatial resolutions of five meters and below is constrained in urban settings because of limitations due to the mixed pixel information. This study investigates the potential of Sentinel-2 products for the detection of roof materials in the city of Kigali as input for planning decisions and monitoring of urban expansion. A linear spectral unmixing is applied to manually collected spectral information of both artificial and natural surfaces. Results indicate that both spatial and radiometric resolution are sufficient to detect even small fractions of roof materials, although a proper calibration and endmember optimization is required. Challenges remain regarding surfaces of similar spectral signatures, such as clay roofs and argillaceous soil, and surfaces which are strongly under-represented within pixels. The spatial distribution of clay tiles which are an indicator for upper-class houses, matches the reference data collected in the fields, but more attention has to be placed on metal roofs of different colors for a reliable prediction of all materials.

* Benjamin Beaumont, Nathalie Stephenne, Laura Van de Vyvere and Eric Hallot. Users’ consultation process in building a land cover and land use database for the official Walloon Georeferential
Abstract

INSPIRE and PSI directives induced major changes in the Walloon geomatics ecosystem. Namely, Wallonia proposed to develop a Georeferential aiming at providing free, open-access and consistent reference geodatasets to the Walloon stakeholders. Some of the current data being outdated, the Georeferential proposes pilot projects for a collaborative and user-oriented design of new maps. One of the categories being identified for the Georeferential is land cover and land use. Build on the knowledge of past researches, the Walous project aims at combining very high spatial with very high temporal resolution Earth observation and thematic data in a classification scheme. To ensure that the new data covers the general interest and to ensure his appropriation by the users, gathering requirements is one of the most vital pieces. Poor gathering of user needs could lead to an unused final product. Walous carried out interviews for clarifying data requirements. This paper introduces this methodology and illustrates how the results impacted the priorities and the specifications of the land cover and land use data.

* Wen Liu, Fumio Yamazaki and Yoshihisa Maruyama. Extraction of the inundated area due to the July 2018 Western Japan torrential rain using multi-temporal ALOS-2 data
Abstract

Successive heavy rainfall affected the western Japan from the late June to the early July 2018. Increased river water overflowed and destroyed river banks, which caused flooding in vast areas. In this study, two pre-event and one co-event ALOS-2 PALSAR-2 images were used to extract inundation areas in Kurashiki and Okayama Cities, Okayama Prefecture, Japan. First, the difference between the pre-event and co-event coherence values was calculated. The decreased coherence areas were extracted as possible inundation. Then the water regions were extracted by the threshold values from the three-temporal intensity images. The increased water regions in July 2018 were obtained as inundation. Finally, the extracted results from the coherence and intensity images were merged to create an inundation map. The results were verified by comparing with a web-based questioner survey report and visual interpretation of aerial photos.

* Stefanos Georganos, Tais Grippa, Assane Niang Gadiaga, Sabine Vanhuysse, Stamatis Kalogirou, Moritz Lennert and Catherine Linard. An Application of Geographical Random Forests for Population Estimation in Dakar, Senegal using Very-High-Resolution Satellite Imagery
Abstract

In this paper we investigate a local implementation of Random Forest (RF), named Geographical Random Forest (GRF) to predict population counts with Very-High-Resolution Remote Sensing (VHHRS) data. As an independent variable we use population density at the neighborhood level from the 2013 census of Dakar, while as explanatory features, the proportions of three different built-up types in each neighborhood derived from a VHHRS land cover classification. The results demonstrated, that by using an appropriate geographic scale to calibrate GRF, we can maximize prediction accuracy due to the incorporation of spatial heterogeneity in the estimates. Additionally, since GRF is an ensemble of local sub-models, the results can be mapped, highlighting local model performance and other interesting spatial variations. Consequently, GRF is suggested as an interesting exploratory and explanatory technique to model remotely-sensed spatially heterogeneous relationships.

* Aurélie Michel, Laure Roupioz and Xavier Briottet. Land Surface Temperature Retrieval over Urban Areas from simulated TRISHNA data
Abstract

The future space joint-mission TRISHNA (Thermal infraRed Imaging Satellite for High-resolution Natural resource Assessment, CNES and ISRO) will allow to retrieve LSTs at 60-m spatial resolution every 3 days, improving the urban environment monitoring capacities. Two methods can be used to derive LST from remote sensing TIR data, Split-Window (SW) or TES (Temperature Emissivity Separation). Even if proven efficient, the complexity and heterogeneity of urban areas limit the performance of those methods. This paper focuses on the TES algorithm. For TES, errors are generated by the impact of the surface roughness on the data. In order to investigate the efficiency of the future TRISHNA mission, research was conducted to evaluate the TES algorithm. To do so, the impacts of using a representative material database when using TES were explored and validated based on airborne or satellite data, and the impact of the surface geometry on the TES estimations was evaluated using 3D thermo-radiative model simulations. This study shows that TES can be used for TRISHNA images with a mean difference of 1.6 K between validation measurements and the retrieved LST for both daytime and nighttime images and that neglecting the impact of the surface geometry while performing TES could lead to errors up to 4 ◦ C in LST estimations.

* Ryan Engstrom, Dan Pavelesku, Tomomi Tanaka and Ayago Wambile. Mapping Poverty and Slums Using Multiple Methodologies in Accra, Ghana
Abstract

Providing housing to slum dwellers, protecting them from natural disasters and diseases, and connecting them to jobs and services through improved infrastructure are urgent policy issues in many Sub-Saharan African cities. Identifying the location and living conditions of slums is a critical step toward designing effective urban policies. By combining household survey data and census data with high spatial resolution satellite imagery and other geospatial data using multiple methodologies, including machine learning, we attempt to define slums objectively within the city of Accra. Within these defined slum areas, the patterns of monetary poverty are assessed. Poverty rates are estimated at the neighborhood level and indicate that living in slums is strongly correlated with higher monetary poverty. Poverty is more prevalent in communities in areas of lower elevation, which in Accra are generally flood-prone areas. However, the results also suggest that not all people living in slums are living in monetary poverty. These results have important policy implications and are crucial to how economic opportunities are generated in slums so that effective urban policies can be designed.

* Oladimeji Mudele and Paolo Gamba. Mapping vegetation in urban areas using Sentinel-2
Abstract

The rapid expansion of cities globally leads to new challenges related to quality of life and health. The presence and fractional distribution of vegetation within urban cities directly impact the life and health of urban dwellers. This paper presents an approach to map urban vegetation from Sentinel-2 data. The twin Sentinel-2 satellites offer a 5-day revisit time global coverage at unprecedented spatial and temporal resolution. The temporal resolution allows for seasonal aggregation of the input data, thus providing phenological information. By considering seasonally aggregated Normalized Difference Spectral Vector (NDSV), a classification was performed using Random Forest (RF) and compared with Classification and Regression Trees (CART) and Support Vector Machines (SVM).

* Xiangli Yang, Loïc Denis, Florence Tupin and Wen Yang. SAR Image Despeckling Using Pre-trained Convolutional Neural Network Models
Abstract

Despeckling is a longstanding topic in synthetic aperture radar (SAR) images. Many different schemes have been proposed for the restoration of SAR images. Among the different possible strategies, the methods based on convolutional neural networks (CNNs) have shown to produce state-of-the-art results on SAR image restoration. However, to learn an effective model is necessary for training a large number of suitable SAR images. To handle this problem, we propose to directly use the pre-trained CNN models on additive white Gaussian noise (AWGN) and transfer them to process SAR speckle. To include such CNNs Gaussian denoisers, we use a multi-channel logarithm approach with Gaussian denoising (MuLoG). Experimental results, both on synthetic and real SAR data, show the method to achieve good performance.

* Benjamin Bechtel, Panagiotis Sismanidis, Wenfeng Zhan and James Voogt. Seasonal Surface Urban Heat Island Analysis
Abstract

Surface Urban Heat Islands (SUHIs) show large temporal variability, which introduces considerable uncertainty into studies based on a single or a few acquisitions. Besides short-term and more random influences, they exhibit considerable systematic diurnal and seasonal variation, which has been investigated in several recent studies. Here an enhanced method for seasonal SUHI analysis based on time series analysis is presented and tested in the southwest United States using MODIS day- and nighttime land surface temperature data. Despite large scatter at the daily time scale, the method is able to identify and parameterize seasonal signals in time series. The seasonal patterns vary substantially between cities. Coastal cities show the largest nighttime SUHI intensity in spring and the lowest in late summer, while continental cities have their SUHI intensity peaks in September and October. During summer, some cities show a daytime cool island. Our study underlines the need to consider temporal patterns in SUHI assessment. The method proved useful to investigate the SUHI seasonality.

* Ryan Engstrom, Robert Harrison, Michael Mann and Amanda Fletcher. Evaluating the Relationship Between Contextual Features Derived from Very High Spatial Resolution Imagery and Urban Attributes: A Case Study in Sri Lanka
Abstract

Extracting information about variations with urban areas satellite imagery has generally focused on mapping individual buildings or slum versus non-slum areas. While these data are useful they can run into issues in very dense urban areas, additionally slums have a subjective definition. In previous research we have found that contextual features are related to population, census variables, poverty, and other values, but have not explored which urban attributes (i.e., buildings and roads) these features represent. In this study we seek to determine the correlation between contextual features calculated on Very High Resolution (VHR) satellite data and urban attributes derived from Open Street Map (OSM) for portions of multiple cities in Sri Lanka. Results indicate that individual contextual features are highly correlated with building area, building density, road area, road density, total built up areas and other features. Moreover, when multiple contextual features are used within a model they can explain from 70 to 97 percent of the variance of these urban features within the study area. This indicates that contextual features are very strong indicators of urban variability and can be used to map differences within the urban setting. This may allow us to forgo having to map each building and road individually for mapping urban areas in future projects.

* John Friesen, Michaela Lestakova, Jens Kaltenmorgen, Suwaythan Nahaganeshan, Peter F. Pelz, Hannes Taubenböck and Michael Wurm. Size Distributions for Morphological Slums in Asia and South America
Abstract

The size distribution of cities within countries was investigated for several years, leading to the famous Zipf’s law. The question arises, if there were similarities between city size distributions across countries and the distribution of a specific urban class: slums. We investigate the size distribution of morphological slums classified by using remote sensing data in three Asian and three South American cities using three different distribution functions (Generalized Pareto Distribution, Lognormal Distribution, and Double Pareto Lognormal Distribution) which were used in the context of inter urban size distributions before. We applied different goodness of fit tests showing that the Double Pareto Lognormal is the best function to both investigated global regions.We find that intra-urban size distributions of slums can be described with similar distribution functions as city size distributions across countries.

* John Friesen, Christoph Knoche, Jakob Hartig, Peter Pelz, Hannes Taubenböck and Michael Wurm. Sensitivity of slum size distributions as a function of spatial parameters for slum classification
Abstract

In a recent work it was shown that the size distribution of slums seems to be similar independent of city, country and continent. However, parameter defining the measurement of slum sizes, i.e. what distance between two slum areas is used to count them as two separate areas, has not been investigated. The present work finds that the similar size of slums of about 104 m2 is independent from the separation. On closer examination, the distance of separation is not sufficient for the partitioning of slums. In addition to the distance, the type of land use should be considered in partitioning algorithms and the planning of infrastructure for slums.

* Chunping Qiu, Michael Schmitt, Hannes Taubenböck and Xiao Xiang Zhu. Mapping Human Settlements with Sentinel-2 Imagery and Deep Neural Networks
Abstract

This paper explores the potential of multi-spectral Sentinel-2 imagery for human settlement mapping, using deep learning based methods. We show the first results of a study area in central Europe, with both a basic residual neural network and a specifically adapted version to better exploit the spectral information. Based on the results and comparison with the existing products, we discuss two interesting questions: how can human settlement mapping be made consistent with or complementary to the existing human settlement maps and how can further improvement in human settlement mapping be achieved?

* Wieke Heldens, Björn Maronga, Julian Zeidler, Farah Kanani-Sühring, Wiebke Hanke and Thomas Esch. Remote sensing-supported generation of surface descriptors for a highly detailed urban climate model
Abstract

Urban climate models become increasingly important for sustainable urban development. The turbulence-resolving micro climate model PALM-4U is currently being developed to better meet the high demands of urban planning related to the spatial detail of modeling scenarios. The study presented here, shows which data sources and approaches can be used to supply the model with the required highly detailed urban surface layers. – It is thereby jointly made use of remote sensing, municipal and open data to generate surface layers at 1m spatial resolution. The surface layers describe buildings, vegetation, terrain, water, pavement, streets and bridges. The heterogeneous data sources and their differing quality and standards require the utilization of various remote sensing and GIS techniques. In this study it is demonstrated that the surface layers can successfully be for urban climate simulations. To ensure easy transferability, itshould be considered to use thematically and/or coarser but more standardized data, after a sensitivity analysis of PALM-4U

* Sara Top, Steven Caluwaerts, Bert Van Schaeybroeck, Rafiq Hamdi, Francois Duchene and Piet Termonia. Modelling the urban heat island: sensitivity to land cover data
Abstract

The MOCCA (MOnitoring the City’s Climate and Atmosphere) network is measuring since July 2016 the urban climate of Ghent by using high-accuracy weather stations installed in different urban environments. Urban canopy temperature observations clearly show that surface properties (e.g. land cover fraction, building height,…) determine the urban heat island (UHI) intensity. This relationship is also expected in simulations of the UHI since the radiation and energy balance in a land surface model, coupled to an atmospheric model, is based upon the characteristics of the surface. In this study we illustrate how the use of more accurate land cover data in ALARO model runs coupled to the SURFEX land-atmosphere interface leads to an improved simulation of the nocturnal UHI of Ghent. As the modelled UHI is very sensitive to the land cover fractions, it can be concluded that it is important to use accurate, up-to-date surface information for UHI modelling.

* Clément Rambour, Loïc Denis and Florence Tupin. Urban surface recovery through graph-cuts iver SAR tomographic reconstruction
Abstract

The understanding of tomograms can be a difficult task when the observed scene is complex. The Synthetic Aperture RADAR (SAR) ranging acquisition induces geometrical distortions depending on the local slope of the back-scattering surface. Over dense urban areas, this particular geometry produces tilted facades instead of straight buildings. Moreover, the high dynamic in SAR images of urban areas may lead to 3-D representation with very bright voxels eclipsing other echoes.

In this paper we present a graph-cut approach adapted to any tomographic reconstruction technique to segment the urban surface in a 3-D tomographic reconstruction. Results on real data from a stack of 40 TerraSAR-X images are presented.

* Tais Grippa, Stefanos Georganos, Sabine Vanhuysse, Moritz Lennert, Nicholus Mboga and Eléonore Wolff. Mapping slums and model population density using earth observation data and open source solutions
Abstract

This paper presents a complete framework for mapping land cover, land use and estimate population densities from very-high resolution images. It proved its ability to accurately extract slums location and extent from the rest of the city. Moreover, the processing chain developed are able to deal with large amount of data and then produce those information citywide.

* Monika Kuffer, Claudio Persello, Karin Pfeffer, Richard Sliuzas and Vinodkumar Rao. Do we underestimate the slum population?
Abstract

According to UN-Habitat, around one billion people live in slum conditions, this figure is reported for the SDG indicator 11.1.1 (the proportion of urban population living in slums, informal settlements or inadequate housing). However, this number comes with many uncertainties. For several countries, estimates are not available, while for other countries reported data might not reflect the real population living in slum conditions. This paper uses Dar es Salaam in Tanzania, as a showcase on how a combination of data extracted from remote sensing combined with locally available sample data and non-official data (e.g., from NGOs) could allow quantifying the degree of uncertainty about city-level slum population estimates. For the city of Dar es Salaam the estimates based on the census data indicate that around 3 million of its inhabitants are living in slum-like conditions, while using a combination of household surveys, settlement level estimates from Slum Dweller International combined with rooftop outlines extracted from unmanned aerial vehicle (UAV) images, the estimated slum population is around 5 million. This raises the question of how much on a global level do we underestimate the number of people living in slum conditions and shows the potential of remote sensing to shad a light on this neglected issue.

* Stenka Vulova and Birgit Kleinschmit. Thermal behavior and its seasonal and diurnal variability of urban green infrastructure in a mid-latitude city – Berlin
Abstract

Urban green spaces refer to a variety of vegetated open areas in cities, including public parks, residential gardens, green roofs, and street trees [7]. The benefits of urban green spaces have been well-documented, with effects on air temperature, air quality, biodiversity, building energy consumption, and soil stabilization [7]. Land surface temperature (LST) is a key variable in characterizing the surface energy and water balance at the land surface-atmosphere interface [1, 17]. Vegetation reduces the LST by providing shade and absorbing radiation energy via transpiration and photosynthesis [5, 16]. In urban environments, vegetation is generally spatio-temporally heterogeneous, with variations in vegetation type, species, vegetation density, vegetation height, leaf area, microclimate, water accessibility, and soil and water characteristics [7]. A review of research on urban green spaces recommended that future research should incorporate functional, structural, and configurational parameters of urban vegetation in order to more fully assess the thermal effect of green spaces [3]. This study will help fill this research gap by considering satellite-derived LST (Landsat 8 and MODIS), 3D vegetation data, biotope types, and vegetation indices in a study of LST across diurnal and seasonal temporal scales in a mid-latitude city (Berlin) over the length of one year (2018). *[This paper is a preliminary paper submitted in response to a personal invitation by Dr. Ellen Banzhaf to the “Capturing green infrastructure provision using diverse airborne and orbital sensor systems” session].

* Younes Zegaoui, Marc Chaumont, Gérard Subsol, Philippe Borianne and Mustapha Derras. Urban object classification with 3D Deep-Learning
Abstract

Automatic urban object detection remains a challenge for city management. Existing approaches in remote sensing include the use of aerial images or LiDAR in order to map a scene. This is for example the case for patch based detection methods. However these methods do not fully exploit the 3D information given by a LiDAR acquisition because they are similar to depth map. 3D Deep-Learning methods are promising in order to tackle the issue of the urban objects detection inside a LiDAR cloud. Additionally, the acquisition can easily be made with an equipped backpack. In this paper we present result of several experiments on urban object classification with the PointNet network trained with public data and tested on our own data-set. We shows that such a methodology delivers encouraging results, and also identify the limits and the possible improvements.

* Hossein Bagheri, Michael Schmitt and Xiao Xiang Zhu. Towards the Reconstruction of Prismatic Building Models by SAR-Optical Stereogrammetry
Abstract

A huge archive of very high-resolution SAR and optical satellite imagery acquired by different remote sensing satellites provides the opportunity to explore the possibility of 3D-reconstruction by multi-sensor stereogrammetry. This paper investigates the potential of SAR-optical stereogrammetry over urban areas using very-high-resolution imagery acquired by TerraSAR-X and Worldview-2. Furthermore, the potentials and challenges of deriving simple prismatic building models by combining the stereogrammetry results and OpenStreetMap building footprints are discussed. The results of this data fusion research demonstrate the possibility of using SAR-optical stereogrammetry for urban 3D reconstruction at level-of-detail 1

* Thomas Tiessen, John Friesen, Lea Rausch and Peter F. Pelz. Using remote sensing data and cluster algorithms to structure cities
Abstract

The increasing urban population and the resulting lack of reliable water, energy and food supply is a big challenge for cities especially in informal settlements (slums). In order to plan new, better supply structures for cities, it is useful to subdivide the cities into sub structures because ever-growing cities are getting more complex and some analyses cannot be conducted at once with the data for a whole city in a time efficient manner. Therefore, many small, more transparent problems are derived from a large, complex problem. This work explains how these subproblems are generated.

* Filip Sabo, Christina Corbane, Panagiotis Politis, Martino Pesaresi and Thomas Kemper. Update and improvement of the European Settlement Map
Abstract

The revised and enhanced version of the new European Settlement Map is presented together with the first results. An added-value to the previous versions is the improved automatic detection of buildings, automatic extraction of water, extraction of building typology and an information layer allowing to derive city indicators. The ESM workflow is fully automatic and runs on the Joint Research Centre Earth Observation Data and Processing Platform.

* Qunshan Zhao, Ryan Reynolds, Chuyuan Wang and Elizabeth Wentz. A multidimensional urban land cover change analysis in Tempe, AZ
Abstract

Rapid population growth leading to significant conversion of rural to urban lands requires deep understanding on how the human population interacts with the built-environment. Our research goal is to explore methodologies on how to analyze multidimensional urban change with the consideration of time, space, and landscape patterns. Using NAIP high resolution satellite images and LIDAR data, we were able to derive land cover classification maps and normalized height difference at different time periods. Then we performed the 2D, 3D and landscape pattern change analysis for a case study area. The research results show that a combination of 2D, 3D and landscape pattern change analysis can provide a comprehensive understanding of urban change, and the results will help urban planners and decision makers to better understand the status of urban transformation and design city for the future.

* Jiameng Lai, Wenfeng Zhan and Sida Jiang. Forecasting of the Nighttime Surface Urban Heat Islands under Clear-sky
Abstract

Modelling of the Surface Urban Heat Island (SUHI) temporal variations have been a great concern. However, most previous studies only focused on modelling the SUHI variations in the past period, yet those for their future patterns remain rarely investigated. By incorporating various predictable meteorological variables in the SVM regression model, this study achieved an attempt to the prediction for the next-day SUHIs over Chinese main cities. Both the SUHI intensity (SUHII) and the pixel-based Gaussian-simulated LSTs were predicted. The averaged MAE of our predicted SUHII across Chinese megacities is 0.67 K; and the MAE for the LST is generally less than 1.5 K. The incorporation of meteorological variables was shown to greatly contribute to the predicted daily SUHIIs. We consider our study, by achieving an attempt to the SUHI prediction, can improve the understanding of the SUHI mitigation.

* Mengbin Rao, Liang Tang, Ping Tang and Zheng Zhang. ES-CNN: An End-to-End Siamese Convolutional Neural Network for Hyperspectral Image Classification
Abstract

In recent years, deep learning-based methods have achieved great success in remote sensing image analysis. However, especially in the context of hyperspectral image classification, there is still a lack of labelled samples to feed those data-hungry deep models. To augment the amount of input data, models operate on pixel-pairs have been proposed and Siamese convolutional neural network (S-CNN) is a typical one. S-CNN is used as a pixel-pair feature extractor and an additional classifier like SVM is required. In this paper, we propose an end-to-end version of S-CNN. Taking advantage of the pairwise input and to make better use of spatial information, a voting strategy using neighboring pixels is also employed to determine the final class label of the center pixel. Experimental results on real hyperspectral datasets show that the proposed method outperforms the original S-CNN by a considerable margin.

* Flora Weissgerber, Elise Colin-Koeniguer, Jean-Marie Nicolas and Nicolas Trouvé. Evolution of the polarimetric behaviour of urban areas with the resolution of SAR images
Abstract

The refinement of PolSAR image resolutions enables to tackle new applications for monitoring of urban areas. Never- theless, this resolution change can have an impact on other image characteristics, such as the polarimetric response. This paper studies the modification of PolSAR urban image statistics due to resolution improvement, by comparing the behviour of clutter patches and built-up areas. Our dataset includes images which resolution has been computationnaly coarsened and PolSAR data acquired at different resolutions. On vegetated clutter, we show that the degree of coherence between HH and VV decreases with the resolution. On build-up areas, we show that he polarimetric entropy becomes more contrasted when the resolution increases. Both results can be explained by an increase of the spatial heterogenity when the resolution is refined.

* Alessandra Budillon, Angel C. Johnsy and Gilda Schirinzi. Contextual Information Based SAR Tomography of Urban Areas
Abstract

SAR Tomography (TomoSAR) is a multidimensional imaging technique that has proven its ability in localizing multiple scatterers in the three dimensional observed scene, allowing the reconstruction of the elevation profile of the structures on the ground. Tomographic approaches usually estimate the elevation distribution of the scetterers in each range-azimuth pixel independently from the neighboring ones (local approaches). Then, any relation among the elevations neighboring pixels is imposed in the tomographic processing. In this paper a local contextual information contained in the data is exploited with the aim of improving the 3D reconstruction (semi-local approaches). Results on real data validate the proposed approach.

* Eleanor Stokes, Miguel Roman, Zhuosen Wang, Ranjay Shrethsa and Tian Yao. URBAN APPLICATIONS OF NASA’s BLACK MARBLE PRODUCT SUITE
Abstract

The Suomi National Polar-Orbiting Partnership (Suomi-NPP) Visible Infrared Imaging Radiometer Suite (VIIRS) Day Night Band (DNB) provides global measurements of nighttime visible and near-infrared light that opens up new opportunities for urban research. While nighttime satellite imagery from the Defense Meteorological Satellite Program Operational Linescan System (DMSP-OLS) heritage product has been used extensively for urban applications since the 1990s, many studies have been limited by its coarse spatial and radiometric resolution, and lack of calibration across time. As a result, urban studies most often used nighttime lights to differentiate between urban and non-urban land, or as a correlate to other variables of interest.
To characterize within-urban land patterns and activities, urban infrastructure dynamics, and small human settlements, science-quality products were needed.Since the first images from VIIRS DNB became available in early 2012, NASA has worked to develop a new set of nightlight products from VIIRS, called the Black Marble Product Suite [1] . Black Marble maximizes the capabilities of VIIRS DNB to generate new information useful for understanding urbanization processes, urban functions, and the vulnerability of urban areas to hazards. In a 2015 case study, our team demonstrated that tracking daily dynamic nighttime measurements from Black Marble can provide valuable information about the character of urban activities and behaviors that shape energy infrastructure and use [2]. Recent work has focused on the ability of Black Marble to track impacts and recovery of the electricity grid after disasters, such as in the case of Hurricane Maria in Puerto Rico, as well as electrification in Ivory Coast, Africa. As the time series expands and is fused with other sources of spatial data, Black Marble has the potential to increase our understanding of changes in urban infrastructure and human factors that influence a myriad of urban sustainability outcomes—including urban resilience.

* Martin Weinmann and Michael Weinmann. Urban Scene Labeling Based on Multi-Modal Data Acquired from Aerial Sensor Platforms
Abstract

In this paper, we address urban scene interpretation on the basis of multi-modal data acquired from aerial sensor platforms. These data comprise RGB color information, hyperspectral information and 3D shape information. As hyperspectral data is known to contain a high degree of redundancy which, in turn, might affect the quality of derived classification results, we also involve techniques for dimensionality reduction and feature selection as well as a transformation of hyperspectral data to high-resolution multispectral Sentinel-2-like data. We use the different types of data to define sets of radiometric and geometric features which are provided separately and in different combinations as input to a Random Forest classifier. To assess the potential of the different types of data and their combination for urban scene interpretation, we present results achieved for the MUUFL Gulfport Hyperspectral and LiDAR Airborne Data Set.

* Nicholus Mboga, Stefanos Georganos, Tais Grippa, Moritz Lennert, Sabine Vanhuysse and Eléonore Wolff. Weakly supervised fully convolutional networks using OBIA classification output
Abstract

In this paper, we present a methodology for preparing reference data to be used in training a fully convolutional neural network. It is a laborious and time-consuming task to prepare adequate training data for a Fully convolutional network (FCN) since the tiles need to be fully labeled. Weakly supervised learning is used when there are inadequate, inaccurate or incomplete training labels. In this paper, an existing semi-automatic Object-based image analysis (OBIA) chain is used to classify a very high-resolution aerial (VHR) imagery of the city of Goma, the Democratic Republic of Congo. Afterwards, fully labeled training tiles are automatically extracted and used to train an FCN designed with a skip architecture and dilated convolutions. An overall accuracy of 91.14% is attained from the preliminary tests, which demonstrates that FCN is robust to noisy labels. Future steps will entail the evaluation of ensemble voting and class probabilities in preparing the training data. This approach is promising and can address the challenge of preparing a large amount of training data for training FCNs.

* Yao Sun, Yuansheng Hua, Lichao Mou and Xiao Xiang Zhu. Large-scale Building Height Estimation from Single VHR SAR image Using Fully Convolutional Network and GIS building footprints
Abstract

Height reconstruction of large-scale buildings from single very high resolution (VHR) SAR image is of great interest especially in applications with temporal restrictions. The problem is highly challenging due to the inherent complexity of SAR images, e.g., side-looking geometry and different microwave scattering contributions. In this work, we present a framework to estimate large-scale building heights from single VHR SAR image. The individual buildings are defined by GIS data, and deep neural network is used to segment wall area in SAR image. The wall layover length is then converted to height and assigned to each building footprint. Experiment in center Berlin area shows results of overall instance height accuracy around 2.54 meters.

* Hossein Aghabababei, Giampaolo Ferraioli and Gilda Schirinzi. Toward 3D Classification of Urban Area With Polarimetric SAR Tomography
Abstract

The classification of urban areas in terms of identification of targets with similar backscattering but having different height position (f.i. road and roof) is still a challenging task. The synthetic aperture radar (SAR) tomography (TomoSAR) approaches estimate scene reflectivity along elevation coordinate, that brings motivation to employ these techniques in the generation of 3D land use map of observed scene. To this aim, the main purpose of this paper is to investigate how TomoSAR can be employed to distinguish different targets by taking the advantage of polarization diversity. In order to fully characterize the backscattering power from targets, and based on the polarimetric basis transformation, the concept of polarization signature is extended to TomoSAR, which typically allows construction of a cube of information with different reflectivity profiles at different polarizations. Hence, on the basis of reference targets, a knowledge-based classification can be used to classify the pixels of image. The framework is evaluated using fully polarimetric multi-baseline data by DLR’s ESAR system over the city of Dresden

* Matthew Gibson, Dhruv Kaushik and Arcot Sowmya. Robust CNNs for detecting collapsed buildings with crowd-sourced data
Abstract

Wildfires are increasingly common and responsible for widespread property damage and loss of life. Rapid and accurate identification of damage to buildings and other infrastructure can heavily affect the efficacy of disaster response during and after a wildfire. We have developed a dataset and a convolutional neural network-based object detection model for rapid identification of collapsed buildings from aerial imagery. We show that a baseline model built with crowd-sourced data can achieve better-than-chance mean average precision of 0.642, which can be further improved to 0.733 by constructing a new, more robust loss function.

* Lakshya Garg, Sandeep Singh, Vaishangi Bajpai and Parul Shukla. Pixel Level Segmentation for Land Use Land Cover Classification using mUnet: A modified Unet Architecture
Abstract

Land-use-land-cover classification(LULC) is used to automate the process of providing labels, describing the physical land type to represent how a land area is being used. Many sectors such as telecoms, utility, hydrology etc need land use and land cover information from remote sensing images. This information provides an insight into the type of geographical distribution of a region with providing low level features such as amount of vegetation, building area, and geometry etc as well as higher level concepts such as land use classes. This information is particularly useful for resource-starved rapidly developing cities for urban planning and resource management. In this paper, we analyze patterns of land use in urban and rural neighborhoods using high resolution satellite imagery, utilizing a state of the art deep convolutional neural network. The proposed LULC network, termed as mUnet is based on an encoder-decoder convolutional architecture for pixel-level semantic segmentation. We test our approach on 3 band, FCC satellite imagery covering 225 km2 area of Karachi. Experimental results show the superiority of our proposed network architecture vis-a-vis other state of the art networks.

* Giampaolo Ferraioli, Vito Pascazio and Sergio Vitale. A Novel Cost Function for Despeckling using Convolutional Neural Networks
Abstract

Removing speckle noise from SAR images is still an open issue. It is well know that the interpretation of SAR images is very challenging and despeckling algorithms are necessary to improve the ability of extracting information. An urban environment make this task more heavy due to different structures and to different objects scale. Following the recent spread of deep learning methods related to several remote sensing applications, in this work a convolutional neural networks based algorithm for despeckling is proposed. The network is trained on simulated SAR data. The paper is mainly focused on the implementation of a cost function that take account of both spatial consistency of image and statistical properties of noise.

* Carlos de Wasseige and Annemarie Enk. Identification of appropriate data sources and analysis software to monitor the growth of informal settlements in Namibia
Abstract

One of the main land administration challenges in Namibia concerns urban growth. It is a fast process which is only partly done formally. The aim of this work is to identify and present an optimal prototypical workflow to be used by planning authorities to monitor spatial growth of informal settlements in Namibia. This workflow involves the access to satellite image data sources, processing services for object detection and based upon that for change detection. Very high spatial resolution images have been proven as the most suitable data source. Other sources such as the use of aerial images derived from drones or cadastral data have been looked at but are not applicable to perform informal settlement monitoring on a larger scale and on a regular basis. The current detection practice, from visual interpretation and point location is used as baseline for improvement. Automatic object base detection is promising, but still requires a visual validation and is sensitive to image pair quality. Given all constraints the proposed procedure for an improved operational system includes image differencing and region growing from points.

* Zina Mitraka, Stavros Stagakis, Giannis Lantzanakis, Nektarios Chrysoulakis, Christian Feigenwinter and Sue Grimmond. High spatial and temporal resolution Land Surface Temperature for surface energy fluxes estimation

Abstract

The study of urban climate and the estimation of urban energy fluxes requires frequent and accurate monitoring of land surface temperature (LST), at the local scale. Current and forthcoming space-borne sensors though, do not provide frequent thermal infrared imagery at high spatial and temporal resolution. In this study, a downscaling technique is applied for improving the spatial resolution of thermal infrared observations and the subsequent LST estimation. LST maps of 100 m spatial resolution multiple times per day were derived for three cities, London, United Kingdom, Basel, Switzerland and Heraklion, Greece. The maps were validated using radiation measurements from flux towers. The accuracy of the derived maps was also assessed using LST data from ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer).

* Marie Jagaille, Nicolas Bellec and René Garello. Regional Copernicus: foster end users uptake by co-designing regional EO-based products and services. An example with local and regional authorities (LRAs) in Brittany
Abstract

Regional level is relevant to address user uptake of Copernicus-based products and services. The Regional Copernicus aims at providing local and regional authorities a framework and a support to become active agents of earth observation-based innovation. It relies on geomatics community, already well-structured in Brittany through the GéoBretagne partnership. Expected result of this approach is to produce and make available new Copernicus-based products and services, co-designed by all the actors of the added-value chain and consistent with INSPIRE requirements. For now, the ongoing approach brought together actors of the public sectors and the private sector to build a first regional demonstrator on landcover. In the future other demonstrators should be implemented.

* Yue Huang and Laurent Ferro-Famil. High-Resolution Adaptive 3-D Urban Reconstruction Using Frugal Polarimetric Tomographic SAR Focusing Techniques
Abstract

This paper addresses the 3-D urban reconstruction using a minimal number of acquisitions, i.e. only 3 acquisitions. To overcome the limitations of conventional tomographic techniques encountered in the urban characterization, a polarimetric model-adaptive tomographic estimator, FP-SSF, is proposed, and solved by the proposed polarimetric alternating projection technique. Meanwhile, a new solution for FP-DML technique is derived as a variant, whose performance is evaluated using simulations and compared to the one of FP-SSF estimator. Moreover, the efffectiveness of FP-SSF estimator is demonstrated by using a dual-baseline PolInSAR data set acquired by DLR’s ESAR sensor at L-band over urban areas of Dresden. The results are validated against the ground truth derived from LiDAR data.

* Tatiana Sussel Gonçalves Mendes and Aluir Porfírio Dal Poz. Urban Road Centerline Detection from Classification Framework Integrating RGB Image and LASER Scanning Data
Abstract

Image classification methods have also been used to previously detect roads in complex urban areas that, in some cases, serve as an initial step for more complex tasks regarding the road reconstruction. The refinement of the detected road network has been a common challenge for several researchers, since even integrating additional information, such as proposed in this work often leads to unsatisfactory results, particularly in urban areas, requiring post-processing to obtain a the road network with a maximum completeness and correctness. This work presents a post-processing method that aims to detect the road centerline. Techniques based on perceptual grouping rules (proximity and collinearity) are applied to lines structures derived from road regions, which are obtained from a classified image by Artificial Neural Network (ANN) using as input data RGB aerial and Airborne Laser Scanning (ALS)-derived images. The later ones correspond to the nDSM (normalized Digital Surface Model) image obtained from the laser altimetry data and the intensity image generated from the laser pulse return intensity information. Experimental results showed that the post-processing applied in the road regions allows to reduce the false positives and to obtain road centerline as continuous lines.

* Anna-Maria Bolte and Theo Kötter. Can you see green or blue? On the necessity of visibility analysis of urban open spaces using remote sensing techniques and GIS
Abstract

The ongoing trend towards urbanization and the consequences of socio-demographic and climate change are increasing the pressure on cities worldwide. The planning of urban vegetation and water areas is therefore essential for a sustainable urban development. In addition to a mainly two-dimensional view of urban open spaces, the focus on the visibility of the areas is important for planning as well as the evaluation of the spatial changes in cities. The visibility of urban open spaces supports the health of urban dwellers and their possibilities of participation during urban planning processes. It is relevant for values of real estate or rent at the local level, while at the global level they influence the urban design and promote competition. The visual results could promote a collaboration of different stakeholders as well as enable a spatially and socio equitable planning of urban open spaces. Remote sensing (RS) techniques and Geographic Information Systems (GIS) have proven themselves for applications of visibility analysis in landscape planning and have a high potential for use in urban contexts. However, further developments are necessary to answer in detail interdisciplinary planning questions.

* Divyani Kohli, Monika Kuffer and Caroline M. Gevaert. The Generic Slum Ontology: Can a Global Slum Repository be created?
Abstract

Slums are home to approximately one quarter of the world’s urban population, in most cities of the Global South the majority of the urban population lives in such areas. In support of global slum eradication and transformation policies, such as the SDGs goal 11 that aims to reduce slums by ensuring inclusiveness of urban areas and developments, global consistent information about the amount and spatial distribution of slums across cities in the Global South is needed. We explore the generic slum ontology (GSO) and available spatial data to explore robust and transferable indicators for global slum mapping. The initial results of our analysis demonstrate image features that are potentially useful to describe differences between slum and non-slum built-up areas, but also show the need of local adaptations. In conclusion, this study highlights the opportunities of the GSO for the development of a global slum repository and hence, the importance of the conceptualization of real world features into image domain features.

* Dong Peng, Wen Yang, Xiangli Yang and Heng-Chao Li. Superpixel-Based Urban Change Detection in SAR Images Using Optimal Transport Distance
Abstract

This paper presents a novel method to detect urban changes in synthetic aperture radar (SAR) images by using optimal transport distance (OTD) as the measurement of dissimilarity. First, the multichannel logarithm with Gaussian denoising (MuLoG) scheme is used to efficiently suppress the speckles of the SAR images. The despeckled images are then segmented into superpixels by the simple linear iterative clustering algorithm (SLIC). After that, the difference map is obtained via the use of Gaussian mixture model and OTD. Finally, change detection results are obtained by the generalized Kitler and Illingworth (GKI) thresholding algorithm. Experimental results on TerraSAR-X images over urban areas show the effectiveness of the proposed method.

* Paula Aguirre, Jorge León and Constanza González. Built environment characteristics and disaster risk: a comparison of bushfires in Wye River (Australia) and Santa Olga (Chile) using Sentinel 2 data
Abstract

In this work, multispectral satellite imagery and geographic data are used to quantify parameters of the built environment of two towns devastated by WUI fires in recent years, Wye River in Australia and Santa Olga in Chile. The severity and distribution of fire damage is also characterized from remote sensing data, and correlated with urban morphology parameters in order to identify common vulnerability indicators of settlements exposed to WUI fires. A comparative analysis of the events shows that fire damage is related to proximity to the density of vegetated areas, to the concentration and relative proximity of built elements, and to their materiality and structural design. Hence, urban planning and construction regulations may play a relevant role in the prevention and mitigation of bushfire effects, by regulating the use of different terrain classes, limiting the number and density of buildings in a given area, promoting the use of fire-resistant materials and structures, and planning suppression and defense activities ahead of time.

* Brabant Charlotte, Alvarez-Vanhard Emilien and Houet Thomas. Improving the classification of urban tree diversity from Very High Spatial Resolution hyperspectral images: comparison of multiples techniques
Abstract

The aim of the study is to compare and assess the efficiency of conventional hyperspectral techniques (dimension reduction, learning and classification methods) to classify the urban tree vegetation diversity. A specific focus is made using very high spatial resolution hyperspectral images acquired from airborne sensor and simulated at various spatial/spectral resolutions. Results show that an accuracy of 78.8% can be reached using MNF and SVM methods to classify urban tree diversity. Moreover, results are more sensitive to learning methods rather than dimension reduction or classification ones.

* Adrien Nivaggioli and Hicham Randrianarivo. Weakly Supervised Semantic Segmentation of Satellite Images
Abstract

When one wants to train a network to perform semantic segmentation, creating pixel-level annotations for each of the images in the database is a tedious task. If he works with aerial or satellite images, which are usually very large, it is even worse. With that in mind, we investigate how to use image-level annotations in order to perform semantic segmentation. Image-level annotations are much less expensive to acquire than pixel-level annotations, but we lose a lot of information for the training of the model. From the annotations of the images, the model must find by itself how to classify the different regions of the image. In this work, we use the method proposed by Anh and Kwak [1] to produce pixel-level annotation from image level annotation. We compare the overall quality of our generated dataset with the original dataset. In addition, we propose an adaptation of the AffinityNet that allows us to directly perform a semantic segmentation. Our results show that the generated labels lead to the same performances for the training of several segmentation networks. Also, the quality of semantic segmentation performed directly bythe AffinityNet and the Random Walk is close to the one of the best fully-supervised approaches.

* Arnaud Le Bris and Nesrine Chehata. A comparison of several spectral and spatial configuration for urban material classification
Abstract

Urban material maps are useful for several city modeling or monitoring applications and can be retrieved from remote sensing data.

This study investigates the impact of spectral and spatial sensor configuration on urban material classification results, comparing several configurations corresponding to existing or envisaged airborne or space sensors. Images corresponding to such sensors were simulated out of an airborne hyperspectral acquisition.

At the end, the relevance of an enhanced spectral configuration and especially providing bands from the SWIR domain was proven, as well as the need for a fine spatial resolution to retrieve urban objects. However, the (late) fusion of multispectral imagery at 2 m resolution with hyperspectral data at 8 m resolution was also proven to leads to good results.

* Liu Hao, Luo Jiancheng, Sun Yingwei, Xia Liegang and Hu Xiaodong. Building extraction based on deep learning via very-high-resolution image
Abstract

Urban is the core of human habitation area, and also it is one of the most complex environmental system of the earth’s surface. The classification in cities has always been one of the research hotspots in the field of geoscience. However, the existing algorithms to extract thematic information in urban usually misbehave, especially for the buildings due to the low spatial resolution. In this paper, we proposed a method based on the deep learning technology to learn the features of buildings in Beijing and Langfang with two high spatial resolution datasets, the accuracy of this method could arrive 0.8624 and 0.9517.

* Anne Puissant, Arnaud Sellé, Nicolas Baghdadi, Vincent Thierion, Arnaud Le Bris and Jean-Louis Roujean. The ‘urban’ component of the French Land Data and Services Centre (Theia)
Abstract

The THEIA data and services centre has been created with the objective of increasing the use of space data by the scientific community and the public actors. THEIA structured the French scientific community 1) through a mutualized Service and Data Infrastructure (SDI) distributed between several centers, allowing access to a variety of products; 2) through the setup of Regional Animation Networks (RAN) to federate and animate users (scientists and public / private actors) and 3) through Scientific Expertise Centres (SEC) clustering virtual research groups on a thematic domain. The research works carried out for urban studies in three SEC are presented in this paper. The works are organized around the design and development of value-added products and services.

* Mehdi Maboudi, Jalal Amini and Markus Gerke. Impact of gap filling on quality of road network
Abstract

Many small segments appear on road surface in VHR images. Most road extraction systems have problem in extraction of these small segments and usually they appear as gaps in extracted road networks. However, most approaches skip filling these gaps. This is on account of the fact that usually overall length of the missing parts is very short relative to the total length of the network, which leads to an indiscernible impact of filling these gaps on geometrical quality criteria. In this paper, using two different datasets and a gap-filling approach, we show that utilizing an effective road gap filling can result in a quite tangible topological improvement in the final road network which is highly demanded in many applications.

* Michael Wurm, Matthias Weigand, Thomas Stark, Jan Goebel, Gert G. Wagner and Hannes Taubenböck. Modelling the impact of the urban spatial structure on the choice of residential location using ‘big earth data’ and machine learning
Abstract

People settle in areas of the city which fit to their individual social and economic situation. In consequence, similar social groups can often be found in similar areas of cities – a process commonly known as segregation. These processes are well-studied from a socioeconomic perspective. In this study, in contrast, we address this topic with an explicitly spatial analysis of these living environments. We present an exploratory data analysis approach to study physical characteristics in different living environments based on a large number of variables derived from spatial data such as satellites, OpenStreetMap and statistical data. Several sensitivity analyses are performed to quantitatively analyze the descriptive performance of these spatial variables on three socioeconomic groups: high and low status households as well as the proportion of foreign population. Non-parametric regression models based on random forests yield highest R² of almost 0.52 for the proportion of foreign population.

* Hunsoo Song, Yonghyun Kim and Yongil Kim. Patch-based light convolutional neural network for land-cover mapping using Landsat-8 images
Abstract

Although deep neural networks have yielded promising results in image classification recently, studies aiming at medium-resolution land-cover mapping are limited. This study proposes light convolutional neural network (LCNN) that can efficiently learn not only the spectral information of a pixel but also contextual information by exploiting the information of neighboring pixels. The performance of LCNN was compared to that of deep convolutional neural network, support vector machine (SVM), k-nearest neighbors (KNN), and random forest (RF). SVM, KNN, and RF were experimented with both patch-based and pixel-based system. Three 30-km-by-30-km test sites of the Level II National Land Cover Database (NLCD) were used for a reference map to embrace a wide range of land-cover types, and one Landsat-8 image was used for each test site. To evaluate the performance of classifiers according to the sample sizes, we varied the sample size to three different sizes. The proposed LCNN achieved the highest accuracy 7 out of 9 cases, and the statistical significance of difference to other classifiers was reported. Finally, the computation times of the classifiers were calculated and it has been confirmed that the LCNN has an advantage in large area mapping.

* Xiangtian Yuan, Jiaojiao Tian and Peter Reinartz. Building Change Detection Based on Deep Learning and Belief Function
Abstract

This paper proposes a new approach for building change detection using multi-temporal satellite stereo data. This approach is composed of three main steps. Firstly building probably map can be derived based on the state of art deep learning approach. In the second step, a decision fusion based fusion model is proposed to highlight the building changes from satellite stereo imagery and the digital surface models (DSMs). In the last step, the building probability maps are used in the change fusion model. Experiments on the multi-temporal data acquired over 5 years confirms the effectiveness of the proposed approach.

* Venkata Sai Krishna Vanama and Y. S. Rao. Urban flood mapping of 2016 Bangalore flood event using C-band RISAT-1 SAR images
Abstract

Flood mapping in urban areas is a rigorous and crucial task in disaster management. Bangalore, one of the Indian megacities, has experienced severe flooding in July 2016. To analyze this flood event, RISAT-1 satellite images were acquired before and after the flood. Various change detection methods were applied to the processed SAR images to identify the flood area. Horizontal like polarized data (HH) is highly sensitive to identify permanent water bodies and also flood affected areas. Permanent water bodies and high elevated areas extracted from DEM were masked out form the results for accurate urban flood mapping. The results show that the spatial distribution of flood was better identified by Normalized Change Index (NCI) method. The results reveal that difference and ratio change detection methods ensued in over and underestimation of flood area which may be due to the use of moderate resolution RISAT-1 SAR images. In urban areas, the use of images acquired with RISAT FRS mode may give better results due to its high spatial resolution.

* Mohamed Chelali, Camille Kurtz, Nicole Vincent and Anne Puissant. Urban land cover analysis from satellite image time series based on temporal stability
Abstract

Satellite Image Time Series (SITS) provide valuable information for the automatic mapping of our territories. In this article we focus on the analysis of urban land covers from SITS, trying to evaluate the density of artificialized areas. We hypothesize that such areas do not evolve significantly through time (over the interval of a year) compared to other non-artificialized areas (e.g. agricultural crops, vegetation). The proposed approach is based on a spatio-temporal characteristic measuring the temporal stability of a zone, extracted using the Run Length encoding method. Preliminary results obtained on a series of 41 SENTINEL-2 images highlight the ability of our approach to discriminate different urban land-cover classes (e.g. artificial areas, high density vs. low density housing areas).

* Ronny Hänsch and Olaf Hellwich. Online Random Forests for Urban Area Classification from Polarimetric SAR Images
Abstract

The growing amount of available image data renders methods unfeasible that require offline processing, i.e. the availability of all data in the memory of the computer. This paper illustrates how Random Forests can be trained by batch processing, i.e. at every iteration only a small amount of samples need to be kept in memory. The benefits of this training scheme are illustrated for the use case of urban area detection from PolSAR imagery. The achieved optimization performance is on par with using all data in the standard offline procedure.

* Xiangyu Zhuo, Milena Moenks and Peter Reinartz. Building Semantic Segmentation from Oblique UAV Imagery
Abstract

Building semantic segmentation is a crucial task for building information modeling (BIM). Current research generally exploits terrestrial image data, which provides only limited view of a building. By contrast, oblique UAV imagery can capture richer information of both the building and its surroundings at a larger scale. In this paper, we present a novel pipeline for building semantic segmentation from oblique UAV images using a fully convolutional neural network (FCN). To cope with the lack of UAV image annotations at facade level, we leverage existing ground-view facades databases to simulate various aerial-view images based on estimated homography, yielding abundant synthetic aerial image annotations as training data. The FCN is trained end-to-end and tested on full-tile UAV images. Experiments demonstrate that the incorporation of simulated views can significantly boost the prediction accuracy of the network on UAV images and achieve reasonable segmentation performance.

* Khelifa Djerriri, Rabia Sarah Cheriguene and Dalila Attaf. Extraction of Built-up Areas from Remote Sensing Imagery using One-Class Classification
Abstract

Mapping of built-up areas were always a main concern to researchers in the field of remotely sensing. Thus, several techniques have been proposed to saving technicians from digitizing hundreds of areas by hand. Multiclass classifiers exhibit a very promising performance in terms of classification accuracy. However, they require that all classes in the study area to be labeled. In many applications, users may only be interested in a specific land class. This referred to as one-class classification (OC) problem. In this paper, we compare a binary Support Vector Machine (SVM) classifier, with two OC classifiers, One-Class SVM, and Presence and Background Learning (PBL) framework for the extracting built-up areas from Gaofen-2 and Aster satellites imagery. The obtained classification accuracies show that PBL provides competitive extraction results due to the fact that PBL is a positive-unlabelled method based on neural network in which large amounts of available unlabelled samples is incorporated into the training phase, allowing the classifier to model the built-up class more effectively.

* Cyril Wendl, Diego Marcos and Devis Tuia. Novelty detection in very high resolution urban scenes with Density Forests
Abstract

Uncertainty in deep learning has recently received a lot of attention. While deep neural networks have shown better accuracy than other competing methods in many benchmarks, it has been shown that they may yield wrong predictions with unreasonably high confidence. This has increased the interest in methods that help providing better confidence estimates in neural networks, some using specifically designed architectures with probabilistic building blocks, and others using a standard architecture with an additional confidence estimation step based on its output. This work proposes a confidence estimation method for Convolutional Neural Networks based on fitting a forest of randomized density estimation decision trees to the network activations before the final classification layer and compares it to other confidence estimation methods based on standard architectures. The methods are compared on a semantic labelling dataset with very high resolution satellite imagery. Our results show that methods based on intermediate network activations lead to better confidence estimates in novelty detection, i.e., in the discovery of classes that are not present in the training set.

Index Terms—Uncertainty, Convolutional Neural Networks, Density Forest, novelty detection, land cover

* Florent Guiotte, Sébastien Lefèvre and Thomas Corpetti. Rasterization strategies for airborne LiDAR classification using attribute profiles
Abstract

This paper evaluates rasterization strategies and the benefit of hierarchical representations, in particular attribute profiles, to classify urban scenes issued from multispectral LiDAR acquisitions. In recent years it has been found that rasterized LiDAR provides a reliable source of information on its own or for fusion with multispectral/hyperspectral imagery. However previous works using attribute profiles on LiDAR rely on elevation data only. Our approach focuses on several LiDAR features rasterized with multilevel description to produce precise land cover maps over urban areas. Our experimental results obtained with LiDAR data from university of Houston indicate good classification results for alternative rasters and even more when multilevel image descriptions are used.

* Peijun Li and Wenting Ma. Multi-scale analysis of seasonal changes on nighttime light brightness using monthly VIIRS DNB composites of urban areas in China
Abstract

In this study, seasonal changes in nightlight brightness at different levels were analyzed using monthly VIIRS DNB composites of urban areas in China. The results indicated that VIIRS nightlight brightness values show different seasonal changes at different levels. The results obtained provide useful information for understanding urban environment and spatial variability of its properties.

* Caglayan Tuna, François Merciol and Sébastien Lefèvre. Monitoring Urban Growth with Spatial Filtering of Satellite Image Time Series
Abstract

Monitoring urban growth and change is an important task for urban planning and disaster management. While several change detection approaches have been proposed to deal with growing urban areas, their performances are usually limited due to outliers in Satellite Image Time Series (SITS). In this study, in order to discriminate urban growth from the other changes, we exploit spatial connectivity of the changed pixels. To do so, we first stack SITS to a single synthetic image whose pixel values denote the temporal variability along the series. Then, we propose to rely on efficient and well-established spatial filtering by means of the max-tree image representation, leading to a novel approach for detecting changes in urban areas, and more precisely focusing on the spatial extent of such changes in relationship with the urban growth. Experimental results obtained on Landsat imagery of Dar es Salaam showed that our approach helps to remove outliers from the change map and provides satisfactory accuracy.

* Arnaud Le Bris and Nesrine Chehata. Urban morpho-types classification from SPOT-6/7 imagery and Sentinel-2 time series
Abstract

This paper aims at detecting several urban morpho-type classes out of SPOT-6/7 imagery and Sentinel-2 time series. Urban classes of Urban Atlas are considered. The proposed strategy is a bottom-up one. It first detects basic urban objects (buildings, roads, vegetation), and use them to calculate multi-scale morphological features. These features are then fed to a Random Forest classifier trained from samples out of Urban Atlas urban classes. Obtained results is optionally merged with a Random Forest classification based on Sentinel-2 time series. Obtained results are promising.

* Chanvoleak Ourng, Yvette Vaguet and Anna Derkacheva. Spatio-temporal urban growth pattern in the Arctic: a case study in Surgut, Russia
Abstract

A spatial-temporal analysis has become an essential approach for tracking the trend of change in land use and land cover pattern especially in the context of urbanization. In this paper, we study on the urban land use change over time of an Arctic pioneer town, Surgut (Russia). This town unfold during the soviet era, embedded with oil production, and grew up to become the country oil capital. Surgut is nowadays one of the healthier Russian Arctic towns and has always been attractive even during the severe decrease in population of the Russian Arctic zone over the decade of the USSR collapse. Various multi-temporal optical satellite imageries (1973-2018) were used to generate the map and analyze the urban expansion, land transformation, and growth directions in order to understand the nature of built-up growth in a soviet Arctic city and its post-soviet evolution.

* Felix Bachofer, Thomas Esch, Jakub Balhar, Martin Boettcher, Fabrice Brito, Vaclav Svaton and Marc Paganini. The Urban Thematic Exploitation Platform – Processing, Analysing and Visualization of Heterogeneous Data for Urban Applications
Abstract

Urbanization is among the most relevant global trends related to the human presence on the planet. As such, it poses major challenge for the well-being of the next generation. To fully understand and properly mitigate the impact of this change, we need precise and up-to-date global monitoring of the urban areas. The TEP Urban platform focuses on delivering multi-source information on trans-sectoral urban challenges. It provides a set of tools for researchers and service providers to process Earth Observation data and integrate them together with data from other sources in order to provide relevant information for spatial planners, policy makers and other stakeholders.