Search

In the context of global change, up-to-date land use land cover (LULC) maps is a major challenge to assess pressures on natural areas. These maps also allow us to assess the evolution of land cover and to quantify changes over time (such as urban sprawl), which is essential for having a precise understanding of a given territory. Few studies have combined information from Sentinel-1 and Sentinel-2 imagery, but merging radar and optical imagery has been shown to have several benefits for a range of study cases, such as semantic segmentation or classification. For this study, we used a newly produced dataset, MultiSenGE, which provides a set of multitemporal and multimodal patches over the Grand-Est region in France. To merge these data, we propose a CNN approach based on spatio-temporal and spatio-spectral feature fusion, ConvLSTM+Inception-S1S2. We used a U-Net base model and ConvLSTM extractor for spatio-temporal features and an inception module for the spatio-spectral features extractor. The results show that describing an overrepresented class is preferable to map urban fabrics (UF). Furthermore, the addition of an Inception module on a date allowing the extraction of spatio-spectral features improves the classification results. Spatio-spectro-temporal method (ConvLSTM+Inception-S1S2) achieves higher global weighted F 1 S c o r e than all other methods tested. [ABSTRACT FROM AUTHOR]

Urban areas are increasing since several years as a result of development of built-up areas, network infrastructure, industrial areas or other built-up areas. This urban sprawl has a considerable impact on natural areas by changing the functioning of ecosystems. Mapping and monitoring Urban Fabrics (UF) is therefore relevant for urban planning and management, risk analysis, human health or biodiversity. For this research, Sentinel-2 (level 2A) single-date images of the East of France, with a high spatial resolution (10 m), are used to assess two semantic segmentation networks (U-Net) that we combined using feature fusion between a from scratch network and a pre-trained network on ImageNet. Moreover three spectral or textural indices have been added to the both networks in order to improve the classification results. The results showed a performance gain for the fusion methods in classifying several UF. However, there is a difference in performance depending on the urbanization gradient; highly urbanized areas provide a better distinction of some UF's classes than rural areas. [ABSTRACT FROM AUTHOR]

Monitoring the spatial and temporal extents of permanent and temporary bodies of surface water is important for various applications such as water resource management, climate modelling, and biodiversity conservation. Satellite remote sensing is an effective source of information to detect surface water over large areas and document their evolution in time. Recently, the European Space Agency (ESA) launched freely available SAR (Synthetic Aperture Radar) and optical sensors (Sentinel-1 & 2) with high revisiting time and spatial resolution. The objective of this paper is to explore the contribution of multi-temporal and multi-source (passive and active) Sentinel observations for improving the detection and mapping of surface waters by applying decision-level image fusion techniques. The approach is tested over Central Ireland using a time series of 16 Sentinel-1 images and a few Sentinel-2 images for the period 2015–2016. Compared to a mono-date approach, the combination of Sentinel-1 & 2 observations provides better accuracy for mapping permanent surface water. Decision level fusion technique allows mapping temporary surface water (such as flooding) with a high accuracy. It also gives the possibility to monitor their dynamics by providing the probability of occurrence of flooded areas at the pixel level. [ABSTRACT FROM AUTHOR]

Reliable information about the spatial distribution of surface waters is critically important in various scientific disciplines. Synthetic Aperture Radar (SAR) is an effective way to detect floods and monitor water bodies over large areas. Sentinel-1 is a new available SAR and its spatial resolution and short temporal baselines have the potential to facilitate the monitoring of surface water changes, which are dynamic in space and time. While several methods and tools for flood detection and surface water extraction already exist, they often comprise a significant manual user interaction and do not specifically target the exploitation of Sentinel-1 data. The existing methods commonly rely on thresholding at the level of individual pixels, ignoring the correlation among nearby pixels. Thus, in this paper, we propose a fully automatic processing chain for rapid flood and surface water mapping with smooth labeling based on Sentinel-1 amplitude data. The method is applied to three different sites submitted to recent flooding events. The quantitative evaluation shows relevant results with overall accuracies of more than 98% and F-measure values ranging from 0.64 to 0.92. These results are encouraging and the first step to proposing operational image chain processing to help end-users quickly map flooding events or surface waters. [ABSTRACT FROM AUTHOR]

This article introduces a new methodology dedicated to classify the evolutions of urban blocks extracted from spatiotemporal topographic databases where an urban block is defined as the smallest area that is surrounded by communication network (roads, railways, . . .). To achieve that, an ascendant hierarchical clustering is applied to sequences of urban block states (i.e., sequences of class labels to which the block belongs to at each date). The principal originality of this approach is to use a distance measure based on DTW (Dynamic Time Warping) which is able to apprehend temporal behaviors (mainly time lags in dates corresponding to a change of state) and which takes into account the semantic proximity between the different kinds of urban blocks. Several experiments have been carried out on areas in the city of Strasbourg (France). First results are relevant and highlight realistic urban dynamics. [ABSTRACT FROM AUTHOR]

Vegetation monitoring is becoming a major issue in the urban environment due to the services they procure and necessitates an accurate and up to date mapping. Very High Resolution satellite images enable a detailed mapping of the urban tree and herbaceous vegetation. Several supervised classifications with statistical learning techniques have provided good results for the detection of urban vegetation but necessitate a large amount of training data. In this context, this study proposes to investigate the performances of different sampling strategies in order to reduce the number of examples needed. Two windows based active learning algorithms from state-of-art are compared to a classical stratified random sampling and a third combining active learning and stratified strategies is proposed. The efficiency of these strategies is evaluated on two medium size French cities, Strasbourg and Rennes, associated to different datasets. Results demonstrate that classical stratified random sampling can in some cases be just as effective as active learning methods and that it should be used more frequently to evaluate new active learning methods. Moreover, the active learning strategies proposed in this work enables to reduce the computational runtime by selecting multiple windows at each iteration without increasing the number of windows needed. [ABSTRACT FROM AUTHOR]

Since vegetation in urban areas delivers crucial ecological services as a support to human well-being and to the urban population in general, its monitoring is a major issue for urban planners. Mapping and monitoring the changes in urban green spaces are important tasks because of their functions such as the management of air, climate and water quality, the reduction of noise, the protection of species and the development of recreational activities. In this context, the objective of this work is to propose a methodology to inventory and map the urban tree spaces from a mono-temporal very high resolution (VHR) optical image using a Random Forest classifier in combination with object-oriented approaches. The methodology is developed and its performance is evaluated on a dataset of the city of Strasbourg (France) for different categories of built-up areas. The results indicate a good accuracy and a high robustness for the classification of the green elements in terms of user and producer accuracies. [ABSTRACT FROM AUTHOR]

The multitude of satellite data products available offers a large choice for urban studies. Urban space is known for its high heterogeneity in structure, shape and materials. To approach this heterogeneity, finding the optimal spatial resolution (OSR) is needed for urban form detection from remote sensing imagery. By applying the local variance method to our datasets (pan-sharpened images), we can identify OSR at two levels of observation: individual urban elements and urban districts in two agglomerations in West Europe (Strasbourg, France) and in Southeast Asia (Da Nang, Vietnam). The OSR corresponds to the minimal variance of largest number of spectral bands. We carry out three categories of interval values of spatial resolutions for identifying OSR: from 0.8 m to 3 m for isolated objects, from 6 m to 8 m for vegetation area and equal or higher than 20 m for urban district. At the urban district level, according to spatial patterns, form, size and material of elements, we propose the range of OSR between 30 m and 40 m for detecting administrative districts, new residential districts and residential discontinuous districts. The detection of industrial districts refers to a coarser OSR from 50 m to 60 m. The residential continuous dense districts effectively need a finer OSR of between 20 m and 30 m for their optimal identification. We also use fractal dimensions to identify the threshold of homogeneity/heterogeneity of urban structure at urban district level. It seems therefore that our approaches are robust and transferable to different urban contexts. [ABSTRACT FROM AUTHOR]

Earth observation data are becoming available at increasingly finer resolutions. Sensors already in existence (IKONOS, Quickbird, SPOT 5, Orbview) or due to be launched in the near future will reach 1-5 m resolution. These very high resolution (VHR) data will provide more details of the urban areas, but it seems evident that they will create additional problems in terms of information extraction using automatic classification. In this framework, this paper examines the potential of the spectral/textural approach to improve the classification accuracy of intra-urban land cover types. The utility of the textural analysis was measured in comparison with multi-spectral per-pixel classifications. Haralick's second-order statistics were applied to the co-occurrence matrix. Four texture indices with six window sizes created from panchromatic images were tested on images at high to very high resolutions (10-1 m). The results show that the optimal index improving the global classification accuracy is the homogeneity measure, with 7 × 7 window size. Moreover, for 1 m images, texture measure of homogeneity allows one to decrease the shadows. [ABSTRACT FROM AUTHOR]

The major part of the population lives in urban areas, and this is expected to increase in the future. The main challenges faced by cities currently and towards the future are the rapid urbanization, the increase in urban temperature and the urban heat island. Mapping and monitoring urban fabric (UF) to analyze the environmental impact of these phenomena is more necessary than ever. This coupled with the increased availability of Earth observation data and their growing temporal capabilities leads us to consider using temporal features for improving land use classification, especially in urban environments where the spectral overlap between classes makes it challenging. Urban land use classification thus remains a central question in remote sensing. Although some research studies have successfully used multi-temporal images such as Landsat-8 or Sentinel-2 to improve land cover classification, urban land use mapping is rarely carried using the temporal dimension. This paper explores the use of Sentinel-2 data in a deep learning framework, by firstly assessing the temporal robustness of four popular fully convolutional neural networks (FCNs) trained over single-date images for the classification of the urban footprint, and secondly, by proposing a multi-temporal FCN. A performance comparison between the proposed framework and a regular FCN is also conducted. In this study, we consider four UF classes typical of many European Western cities. Results show that training the proposed multi-date model on Sentinel 2 multi-temporal data achieved the best results with a Kappa coefficient increase of 2.72% and 6.40%, respectively for continuous UF and industrial facilities. Although a more definitive conclusion requires further testing, first results are promising because they confirm that integrating the temporal dimension with a high spatial resolution into urban land use classification may be a valuable strategy to discriminate among several urban categories. [ABSTRACT FROM AUTHOR]

Image time series, such as Satellite Image Time Series (SITS) or MRI functional sequences in the medical domain, carry both spatial and temporal information. In many pattern recognition applications such as image classification, taking into account such rich information may be crucial and discriminative during the decision making stage. However, the extraction of spatio-temporal features from image time series is difficult to handle due to the complex representation of the data cube. In this paper, we present a strategy based on Random Walk to build a novel segment-based representation of the data, passing from a 2D + t dimension to a 2D one, more easily manipulable and without losing too much spatial information. Such new representation is then used to feed a classical Convolutional Neural Network (CNN) in order to learn spatio-temporal features with only 2D convolutions and to classify image time series data for a particular classification problem. The influence of the way the 2D + t data are represented, as well as the impact of the network architectures on the results, are carefully studied. The interest of this approach is highlighted on a remote sensing application for the classification of complex agricultural crops. [ABSTRACT FROM AUTHOR]

Les cartes d'occupation du sol produites à des résolutions spatiales et temporelles élevées constituent actuellement une ressource très importante pour beaucoup d'organismes privés ou publics. Le développement de méthodes de cartographie automatique, fiables et robustes basées sur la classification d'images satellites constitue ainsi un enjeu majeur. Dans ce cadre, l'imagerie radar apporte l'avantage de fournir des images de jour comme de nuit, et quelles que soient les conditions météorologiques. Plus récemment, l'exploitation des informations de rétrodiffusion fournies par les images SAR (Synthetic Aperture Radar) polarimétriques a permis d'étendre les possibilités apportées par l'imagerie radar. Dans cette étude, une carte d'occupation du sol a été produite sur une partie de la plaine d'Alsace et du massif vosgien à partir (1) de 76 paramètres polarimétriques extraits d'une image ALOS PALSAR en polarisation quadruple et (2) d'une méthode de classification orientée-objet. Plusieurs algorithmes de classification ont été testés et l'algorithme du plus proche voisin est ressorti comme donnant les meilleurs résultats. La méthode mise en place à l'avantage d'être semi-automatique et facilement reproductible. Neuf classes d'occupation du sol ont été cartographiées avec un taux de bon classement de 69%. Plus précisément, trois d'entre elles ont été très correctement détectées : la forêt, l'urbain et l'eau. D'autres classes ont été confondues du fait de la similarité de leur signature polarimétrique : les zones de vignobles, les prairies et les zones de cultures. Enfin, trois classes non visibles sur les données a priori et les images optiques de référence ont pu être identifiées sur l'image polarisée. Ces premiers résultats sont prometteurs pour la cartographie de l'occupation des sols à partir d'images SAR polarimétriques. [ABSTRACT FROM AUTHOR]

This article is concerned with the use of unsupervised methods to process very high resolution satellite images with minimal or little human intervention. In a context where more and more complex and very high resolution satellite images are available, it has become increasingly difficult to propose learning sets for supervised algorithms to process such data and even more complicated to process them manually. Within this context, in this article we propose a fully unsupervised step by step method to process very high resolution images, making it possible to link clusters to the land cover classes of interest. For each step, we discuss the various challenges and state of the art algorithms to make the full process as efficient as possible. In particular, one of the main contributions of this article comes in the form of a multi-scale analysis clustering algorithm that we use during the processing of the image segments. Our proposed methods are tested on a very high resolution image (Pléiades) of the urban area around the French city of Strasbourg and show relevant results at each step of the process. [ABSTRACT FROM AUTHOR]

Au cours de la dernière décennie, la télédétection par imagerie satellitaire a connu plusieurs révolutions, liées d'une part à l'augmentation spectaculaire du nombre de dispositifs d'acquisition, et d'autre part aux progrès des capteurs tant en termes de résolutions spatiale et temporelle que spectrale. Dans ce contexte, de nouvelles problématiques sont apparues, notamment en traitement et analyse d'images. L'un des axes de recherche les plus prometteurs pour traiter ces problématiques s'appuie sur la notion de hiérarchie, qui peut se décliner sous plusieurs formes et ainsi permettre de traiter plusieurs types de tâches, du traitement de bas-niveau jusqu'à l'analyse de haut niveau. Dans cet article, nous décrivons certaines tendances récentes liées à ces approches hiérarchiques. [ABSTRACT FROM AUTHOR]

The widespread introduction of satellite imagery with several spatial, spectral and temporal resolutions is a real opportunity to analyze and characterize the urban environment. These last few years, the Object-Based Image Analysis (OBIA) approach has been largely developed and applied for urban applications. However, a major issue in this approach is domain knowledge formalization and exploitation. In this paper, after a detailed presentation of related works in the domain of semantic interpretation using ontologies, a taxonomy of urban objects and a framework to describe urban objects and their spatial organization is presented. This taxonomy is then formalized in a domain ontology using the semantic web ontology language and rules (OWL and SWRL). [ABSTRACT FROM AUTHOR]

La fusion d'images multispectrales à très haute résolution spatiale (THR) avec des séries temporelles d'images moins résolues spatialement mais comportant plus de bandes spectrales permet d'améliorer la classification de l'occupation du sol. Elle tire en effet le meilleur parti des points forts géométriques et sémantiques de ces deux sources. Ce travail s'intéresse à un processus d'extraction automatique de la tache urbaine fondé sur la fusion tardive de classifications calculées respectivement à partir d'images satellitaires Sentinel-2 et SPOT-6/7. Ces deux sources sont d'abord classées indépendamment selon 5 classes, respectivement par forêts aléatoires et réseaux de neurones convolutifs. Les résultats sont alors fusionnés afin d'extraire les bâtiments le plus finement possible. Cette étape de fusion inclut une fusion au niveau pixellaire suivie d'une étape de régularisation spatiale intégrant un terme lié au contraste de l'image. Le résultat obtenu connaît ensuite une seconde fusion afin d'en déduire une tache urbaine : une mesure a priori de se trouver en zone urbaine est calculée à partir des objets bâtiments détectés précédemment et est fusionnée avec une classification binaire dérivée de la classification originale des données Sentinel-2. [ABSTRACT FROM AUTHOR]

Identification and monitoring of urban fabric and preservation of existing ecosystems have become major issues to maintain or increase biodiversity in areas under urban influence in most of European cities. While many studies have shown the interest of using optical remotely sensed data for that purpose, a consolidated and reproducible methodological framework was still missing. In this context, a multi-scale methodology has been proposed in the framework of the project VALI-URB to map built-up and vegetated land features in urban and suburban areas based on Pléiades images and existing ancillary data (vector databases or the Open Street Map database). The objective of this paper is to highlight the interest of using land cover/use maps derived from Pléiades images and vector databases to semi-automatically characterize grey and green infrastructures at a scale of 1:10,000. First results are presented on two medium-sized cities with different urban forms: Strasbourg and Rennes (France). [ABSTRACT FROM AUTHOR]

Active learning (AL) algorithms have been proven useful in reducing the number of required training samples for remote sensing applications; however, most methods query samples pointwise without considering spatial constraints on their distribution. This may often lead to a spatially dispersed distribution of training points unfavorable for visual image interpretation or field surveys. The aim of this study is to develop region-based AL heuristics to guide user attention toward a limited number of compact spatial batches rather than distributed points. The proposed query functions are based on a tree ensemble classifier and combine criteria of sample uncertainty and diversity to select regions of interest. Class imbalance, which is inherent to many remote sensing applications, is addressed through stratified bootstrap sampling. Empirical tests of the proposed methods are performed with multitemporal and multisensor satellite images capturing, in particular, sites recently affected by large-scale landslide events. The assessment includes an experimental evaluation of the labeling time required by the user and the computational runtime, and a sensitivity analysis of the main algorithm parameters. Region-based heuristics that consider sample uncertainty and diversity are found to outperform pointwise sampling and region-based methods that consider only uncertainty. Reference landslide inventories from five different experts enable a detailed assessment of the spatial distribution of remaining errors and the uncertainty of the reference data. [ABSTRACT FROM AUTHOR]

Abstract: The automated detection and mapping of landslides from Very High Resolution (VHR) images present several challenges related to the heterogeneity of landslide sizes, shapes and soil surface characteristics. However, a common geomorphological characteristic of landslides is to be organized with a series of embedded and scaled features. These properties motivated the use of a multiresolution image analysis approach for their detection. In this work, we propose a hybrid segmentation/classification region-based method, devoted to this specific issue. The method, which uses images of the same area at various spatial resolutions (Medium to Very High Resolution), relies on a recently introduced top-down hierarchical framework. In the specific context of landslide analysis, two main novelties are introduced to enrich this framework. The first novelty consists of using non-spectral information, obtained from Digital Terrain Model (DTM), as a priori knowledge for the guidance of the segmentation/classification process. The second novelty consists of using a new domain adaptation strategy, that allows to reduce the expert’s interaction when handling large image datasets. Experiments performed on satellite images acquired over terrains affected by landslides demonstrate the efficiency of the proposed method with different hierarchical levels of detail addressing various operational needs. [Copyright &y& Elsevier]

Abstract: We propose a new distance called Hierarchical Semantic-Based Distance (HSBD), devoted to the comparison of nominal histograms equipped with a dissimilarity matrix providing the semantic correlations between the bins. The computation of this distance is based on a hierarchical strategy, progressively merging the considered instances (and their bins) according to their semantic proximity. For each level of this hierarchy, a standard bin-to-bin distance is computed between the corresponding pair of histograms. In order to obtain the proposed distance, these bin-to-bin distances are then fused by taking into account the semantic coherency of their associated level. From this modus operandi, the proposed distance can handle histograms which are generally compared thanks to cross-bin distances. It preserves the advantages of such cross-bin distances (namely robustness to histogram translation and histogram bin size issues), while inheriting the low computational cost of bin-to-bin distances. Validations in the context of geographical data classification emphasize the relevance and usefulness of the proposed distance. [Copyright &y& Elsevier]

Abstract: The extraction of urban patterns from very high spatial resolution (VHSR) optical images presents several challenges related to the size, the accuracy and the complexity of the considered data. Based on the availability of several optical images of a same scene at various resolutions (medium, high, and very high spatial resolution), a hierarchical approach is proposed to progressively extract segments of interest from the lowest to the highest resolution data, and then finally determine urban patterns from VHSR images. This approach, inspired by the principle of photo-interpretation, has for purpose to use as much as possible the user''s skills while minimising his/her interaction. In order to do so, at each resolution, an interactive segmentation of one sample region is required for each semantic class of the image. Then, the user''s behaviour is automatically reproduced in the remainder of the image. This process is mainly based on tree-cuts in binary partition trees. Since it strongly relies on user-defined segmentation examples, it can involve only low level—spatial and radiometric—criteria, then enabling fast computation of comprehensive results. Experiments performed on urban images datasets provide satisfactory results which may be further used for classification purpose. [Copyright &y& Elsevier]