different times over urban areas provides meaningful information like the expansion of cities. This paper presents a method for the generation of a coregistered geocoded simulated SARimage using a LiDAR digital surface model (DSM) and the comparison to a real SARimage for the purpose of changedetection. The DSMand metadata of the SARimage are used for the geocoding of the simulated image, while a line extraction and matching algorithm is used to compensate small shifts between the two images. Afterwards the shadow area in the simulated image is extracted and the corresponding mask is used to analyze the real SARimage in order to detect the changes. An application of the proposed concept is presented for the city centre of Munich where the comparison with the TerraSAR-X GEC product shows a good result. The detected positive changes (new buildings) are compared to a 3D change reference map and show a good correspondence.
The colour composition in Fig. 3(a) shows three images over the Chuquicamata copper mine in Chile. The regions of high reflectivity caused by an extreme foreshortening effect are slightly smeared over during the geocoding step. One can note that something has changed around the deposit in the middle left of the image. The other images show the results of a curvelet-basedchangedetection with weighted coefficients. In Fig. 3(b) series of red and green stripes along the moun- tain especially in the upper part are visible. In Fig. 3(c) these structures can be found in the lower part. Fig. 3(d), which depicts the changes between the first and the third image can be seen as sum of the two preceding images. The sequence of red (darkened) and green (brightened up) linear features is most remarkable. These changes indicate a systematic dis- placement of linear high backscatterers. An explanation can be found in the geometrical form of the mine: The deposit is built in terraces. If the deposition goes on, the edges of these terraces move and cause also a displacement of the bright lines in the SARimage, because of the reflector-like diplane backscattering at each stage. In contrast to the colour com- position (Fig. 3(a)) the results are very smooth and show no single pixel changes in the surrounding (Fig. 3(b)-3(d)). So, this approach is capable to survey open cast mining activities even though it is not possible at the moment to determine the amount of soil moved.
In this article, a method based on a non-parametric estimation of the Kullback –Leibler divergence using a local feature space is proposed for synthetic aperture radar (SAR) imagechangedetection. First, local features based on a set of Gabor ﬁlters are extracted from both pre- and post-event images. The distribution of these local features from a local neighbourhood is considered as a statistical representation of the local image information. The Kullback –Leibler divergence as a probabilistic distance is used for measuring the similarity of the two distributions. Nevertheless, it is not trivial to estimate the distribution of a high-dimensional random vector, let alone the comparison of two distributions. Thus, a non-parametric method based on k-nearest neighbour search is proposed to compute the Kullback –Leibler diver- gence between the two distributions. Through experiments, this method is compared with other state-of-the-art methods and the e ﬀectiveness of the proposed method for SARimagechange detec- tion is demonstrated.
Related works reported in the literature mostly compare two SAR images with the same imaging geometry for changedetection. In comparison to LiDAR-SARchangedetection, the SAR-SARchangedetection is free from modeling of real world and assumption of reflection parameters, thus they avoid these modeling errors which may affect changedetection results. However, the traditional SAR-SARchangedetection also has limitations. On one side, the interpretation of the detected changes is difficult, especially for very highresolution data. They may only provide pixels with decreased or increased signals. But the interpretation of these detected changed pixels is still a challenging task. Recently, some articles (Ferro et al. 2013; Marin et al. 2015) have presented methods to solve this problem, but the provided methods can work only for specific buildings (e.g., isolated rectangular buildings). The second limitation is the requirement of SAR images with the same imaging geometry. This requirement might not be fulfilled if we want to use the first available SAR data in crisis situation. For example, the TerraSAR-X satellite has a period of 11 days (which can fulfil the same imaging geometry) and a global access time of maximum three days (which cannot guarantee the same imaging geometry).
PCA techniques were widely used to extract features from different kinds of images. For example,  proposes an unsupervised technique for visual target modeling, which is based on density estimation in high- dimensional spaces using PCA. Such an approach was proved to be well- suited for the detection of facial features. It exploits the redundancy to reduce the dimensionality of the training imagery, in order to form a computationally simple estimator for the complete likelihood function of the object. The estimator in this case, has the advantage that it is based on a subspace decomposition and can be evaluated using only the most principal component vectors. Like human faces, SAR images over urban areas provide a high diversity of features. Thus, a PCA based method should be also adapted to the recognition of the different struc- tures existing in the urban SAR scenes. In , it was demonstrated that the eigenspace relative to the covariance matrix of the training images, provides a well-suited descriptive model of some specific SAR scenes. A PCA is performed on the training set in order to identify the eigen- images that provide the best discrimination between the different classes (called also the eigenspace). This approach was applied to radar target identification in a three-class database formed by tanks, Armored Per- sonnel Carriers (APCs) and self-propelled guns. Such special targets are unfortunately not usual in highresolutionSAR images over urban ar- eas. In general, in urban areas, the mostly found classes consist rather in large/small buildings, vegetation, roads, parking, etc. Processing these classes is much more complex than tanks, APCs and guns, where the distribution variety of the targets is not too large.
provided in literature. This dissertation contributes a pixel-based algorithm to detect increased backscattering in SAR images by analyzing the SAR pixel values according to simulated layers. To detect demolished buildings, simulated images are generated using LiDAR data. Two comparison operators (normalized mutual information and joint histogram slope) are used to compare image patches related to same buildings. An experiment using Munich data has shown that both of them provide an overall accuracy of more than 90%. A combination of these two comparison operators using decision trees improves the result. The fourth objective is to detect changes betweenSAR images acquired with different incidence angles. For this purpose, three algorithms are presented in this dissertation. The first algorithm is a building-level algorithm based on layer fill. Image patches related to the same buildings in the two SAR images are extracted using simulation methods. For each extracted image patch pair, the change ratio based on the fill ratio of building layers is estimated. The change ratio values of all buildings are then classified into two classes using the EM-algorithm. This algorithm works well for buildings with different size and shape in complex urban scenarios. Since the whole building is analyzed as one object, buildings with partly demolished walls may not be detected. Under the same idea, a wall-level changedetection algorithm was developed. Image patches related to the same walls in the two SAR images were extracted and converted to have the same geometry. These converted patch pairs are then compared using change ratios based on fill ratio or fill position. Lastly, the wall change results are fused to provide building change result. Compared to the building-level changedetection algorithm, this method is more time consuming, but yields better results for partly demolished buildings. A combination of these two algorithms is therefore suggested, whereby the building-level method is used for all buildings and wall-level method additionally for selected large buildings. The third developed algorithm is a wall-level changedetection algorithm based on point-feature location. To this end, local maximum points in two SAR images corresponding to the same building façade are compared. This method provides promising result for the present data. It may work better for future data with increased resolution to detect changes of detailed façade structures.
In this paper we present an alternative method for SARimage denoising, structure enhancement, andchangedetectionbased on the curvelet transform. Curvelets can be denoted as a two dimensional further development of the well-known wavelets. The original image is decomposed into linear ridge-like structures, that appear in different scales (longer or shorter structures), directions (orientation of the structure) and locations. The inﬂuence of these single components on the original image is weighted by the corresponding coefﬁcients. By means of these coefﬁcients one has direct access to the linear structures present in the image. To suppress noise in a given SARimage weak structures indicated by low coefﬁcients can be suppressed by setting the corresponding coefﬁcients to zero. To enhance structures only coefﬁcients in the scale of interest are preserved and all others are set to zero. Two same-sized images assumed even a changedetection can be done in the curvelet coefﬁcient domain. The curvelet coefﬁcients of both images are differentiated and manipulated in order to enhance strong and to suppress small scale (pixel-wise) changes. After the inverse curvelet transform the resulting image contains only those structures, that have been chosen via the coefﬁcient manipulation. Our approach is applied to TerraSAR-X HighResolution Spotlight images of the city of Munich. The curvelet transform turns out to be a powerful tool for image enhancement in ﬁne-structured areas, whereas it fails in originally homogeneous areas like grassland. In the changedetection context this method is very sensitive towards changes in structures instead of single pixel or large area changes. Therefore, for purely urban structures or construction sites this method provides excellent and robust results. While this approach runs without any interaction of an operator, the interpretation of the detected changes requires still much knowledge about the underlying objects.
Abstract—This paper presents two changedetection strategies based on the fusion of scene knowledge and two highresolutionSAR images (pre-event, post-event) with focus on individual buildings and facades. Avoiding the dependence of the signal incidence angle, the methods increase the flexibility with respect to near-real-time SARimage analysis after unexpected events. Knowledge of the scene geometry is provided by digital surface models, which are integrated into an automated simulation processing chain. Using strategy 1 (based on building fill ratio; BFR), building changes are detected based on change-ratios considering layover and shadow areas. Strategy 2 (based on wall fill position; WFP) enables one to analyze individual facades of buildings without clear decision from strategy 1, which is based on a geometric projection of facade layover pixels. In a case study (Munich city center), the sensitivity of the changedetection methods is exemplified with respect to destroyed buildings and partly changed buildings. The results confirm the significance of integrating prior knowledge from digital surface models into the analysis of highresolutionSAR images.
of both acquisitions. However, VHR archive SAR imagery acquired shortly before a landslide event, especially at the same image acquisition geometry as the next possible post-event imagery, is in most cases not available. Modern VHR SAR missions, such as TerraSAR-X, COSMO-SkyMed, or RADARSAT-2, do not systematically cover the entire world. Each acquisition has to be programmed manually. Furthermore, due to limited disk space on board the satellites and especially due to limited downlink transmission rates, these sensors are not able to provide worldwide coverage within a short time period—i.e., commonly, no archive image recorded shortly before the event is available. Here, we present a fast and transferable landslide detection methodology based only on post-event VHR PolSAR imagery supported by freely available and systematically-acquired pre-event high-resolution (HR) optical data. The post-event VHR PolSAR acquisition can be programmed before the next overpass of the satellite after the landslide event, independent of any geometrical restrictions by a pre-event SAR imagery. The proposed landslide mapping procedure is a semi-automatic changedetection approach based on pre-event HR optical imagery of Landsat-8 or Sentinel-2 and post-event VHR PolSAR data (e.g., TerraSAR-X) acquired shortly after the event.
multispectral bands. The idea is that the damaged areas should be rich in textural features compared with the same areas after reconstruction which mainly consist of regular buildings. The widely used texture modeling is grey level co-occurrence matrix  due to its simplicity and low computational complexity, which has been proved very efficient in texture modeling. The textural features introduced in this paper are sum variance and difference variance. The details about how to compute them can be found in . As illustrated in Fig. 2, they look considerably different between the two images at different acquisition dates. The texture in QuickBird image looks irregular and coarse, whereas the boudaries of new building can be seen very clearly in GeoEye imagery. All the textural features in Fig. 2 are calculated within 3 × 3 pixel window, which makes a good distinction between the original two images. These two textural bands are then combined with original multispectral images to perform IR-MAD changedetection method described above.
This paper presents a hybrid evolutionary algorithm for fast intensity based matching between satellite imagery from SARand very high-resolution (VHR) optical sensor systems. The precise and accurate co-registration of image time series and images of different sensors is a key task in multi-sensor image processing scenarios. The necessary preprocessing step of image matching and tie-point detection is divided into a search problem and a similarity measurement. Within this paper we evaluate the use of an evolutionary search strategy for establishing the spatial correspondence between satellite imagery of optical and radar sensors. The aim of the proposed algorithm is to decrease the computational costs during the search process by formulating the search as an optimization problem. Based upon the canonical evolutionary algorithm, the proposed algorithm is adapted for SAR/optical imagery intensity based matching. Extensions are drawn using techniques like hybridization (e.g. local search) and others to lower the number of objective function calls and refine the result. The algorithm significantely decreases the computational costs whilst finding the optimal solution in a reliable way.
In this paper, a method for supporting the interpretation of highresolutionSAR images with simulated radar images using a LiDAR DSM has been presented. Line features are extracted from the simulated SARimageand the real SARimageand used for matching of the images. Two subpixel precise translation parameters are calculated from the matching result and enable a direct overlay of all simulated images on the real SARimage. A single building model is generated from the DSMand used for building recognition in SARimage. Special features like double bounce lines, shadow areas in SARimage can be automatically indicated. Future work will concentrate on pixel based analysis of the difference between simulated and real SARimageand derive a semantic description of SAR images.
sual perception of layover varies significantly while the intensity distributions intend to follow log-normal distributions (even for local maxima as a rough approximate). The polarization mode HH dominates VV in intensity (mean, median) but also tends to show stronger intensity variation (standard deviation). In con- trast, no tendency with respect to the proportion between HH and VV is observed on the number of prominent point signatures. As a conclusion, distribution-based methods for layover analysis seem to be favored which may be optionally supported by feature- based concepts, e.g., focused on lines or point-like signatures. The prominent appearance of facade layover motivates the identi- fication of the related SARimage parts in order to extract building- related information. In this context, a concept has been proposed for identifying building layover based on simulation methods in- cluding CityGML data. To this end, a simulation processing chain has been extended to fuse TerraSAR-X images and prior information provided by CityGML data sets. In this context, the data transformation from CityGML to the POV-Ray data struc- ture used by the simulator is fully automated using the spatial ETL (extract, transform, and load) software FME. As a first ex- ample, a case study of the Munich city center has been shown where the extent of simulated building layover is directly super- posed on a geocoded TerraSAR-X image. The simulation concept based on CityGML data indicates that the representation of real world entities by semantic objects has a number of advantages over the purely geometric representation in a DSM. In particular, changedetection applications may benefit from CityGML data sets if they are kept up to date and assigned with meta informa- tion.
Due to the all-weather and all-time data acquisition capability, SAR images are often the only available data in crisis situations, e.g. shortly after an earth- quake. Having one pre-event and one post-event very highresolution (VHR) SARimage, changes within the scene of interest can be recovered  . Howev- er, the change analysis may be hampered by different SAR imaging geometries, missing pre-event SAR da- ta, and the revisit time related to subsequent SAR ac- quisitions. Including a-priori information from other data sources may be helpful in order to resolve some of the limitations, e.g. using optical data .
As shown, the training sample extraction is built on a coarse index-checking procedure. Although we mentioned it is difficult to find indices for urban objects given the varying conditions of the scenes within the time-series, some common indicators can still be used for a coarse estimation. In this procedure, the nDSM (normalized DSM) derived with morphological filtering is used to represent off-terrain objects. MSI (morphological shadow index) and NDVI (normalized difference vegetation index) are used to indicate shadows and vegetation. Although our target is building changedetection, multi-class information can be useful to construct distinctive representation of the building class. Moreover, time series changedetection can be also applied to classes other than buildings. It should be noted that we only update the samples when there is a large difference between these indicators in the time-series. In figure 1, these differencing threshold in our experiments are taken as T1=1.5,
SimGeoI is designed as a framework to relate high- resolution satellite images of SARand optical sensors se- mantically via 3-D space, exploiting geometric a priori scene knowledge. It can thus be a useful tool in solving the alignment problem which is a necessary prerequesite for subsequent SAR-optical data fusion . Related work in the field of remote sensing is mostly concerned with connecting 3-D geometry with either optical or SARimage data. As examples for the first group, changes between digital elevation data and optical satellite data are identified in . Classification methods based on the joint analysis of laserscanning point clouds and optical data are reported in . The impact of scene geometry and sensor perspective has to be accounted for in the field of pan-sharpening where multi-sensor fusion is usually enabled by ortho-correction . In , representing the second group, prominent SARimage signatures (persistent scatterers) are localized at buildings and mapped into slant view optical images for subsequent refinement steps. In the ”SARptical” framework developed by Wang et al. , , a 3-D point cloud, which is generated by co-registration of point clouds derived from SAR tomography and optical dense stereo matching, is used to link SARand optical image data. An example for pioneering work in the field of simulation-basedchangedetection from SARand optical remote sensing data is . The authors use optical data for the manual extraction of building parameters. The resulting 3-D model is thereafter inserted into an iterative SARimagesimulation procedure in order to identify building changes. A fully automated process for scene modeling andsimulation is not yet proposed. Another example is presented by Ali et al., who integrate 3-D city models into the analysis of changes based on SAR images, using optical images for the validation . The impact of geometric projection effects due to object height, however, is not
Image segmentation is the process of partitioning an image into groups of pixels that are spectrally similar. Here, the objective of segmentation is to produce small units that have different spectral characteristics in comparison to the areas nearby. A proper level of segmentation in highresolution satellite images is difficult to reach due to mixed pixels, spectral similarity between different land covers and the textured appearance of specific land covers. Many segmentation methods have been introduced in computer vision, like watershed (Vincent and Soille, 1991), level-set, mean-shift (Comaniciu and Meer, 2002) and several more. All approaches attempt to reach an appropriate segmentation level by adjusting one or more parameters based on one segmentation algorithm (Vincent and Soille, 1991; Meyer and Beucher, 1990; Comaniciu and Meer, 2002; Melendez et al., 2011). Since urban areas typically consist of different land covers (e.g. buildings, roads, shadows, trees), it is difficult to obtain adequate segments for all land covers using a single segment scale, since different kinds of objects require different segmentation levels. Over- and under-segmentation typically appear together. Some methods perform multi-scale segmentation. Since image splitting is more difficult to control, much work has been performed based on merging over-segmentation results (Haris et al., 1998; Ning et al., 2010; Nock and Nielsen, 2004).
In this paper, we worked with TerraSAR-X images. A TerraSAR-X product  is mainly composed of the image data and its metadata. The size of a very highresolution TerraSAR-X image is on average 10000×10000 pixels with varying numbers of bits per pixel(16 or 32), different types of data (float, unsigned int) and consisting of one or multiple bands in GeoTiff format, the corresponding metadata are con- tained in Extensible Markup Language (XML) files containing related information in form of structured text and numbers. An XML product description metadata file comprises about 250 entries grouped into categories such as product compo- nents, product information (i.e., pixel spacing, coordinates, format, etc.), processing parameters, platform data, calibration history, and product quality annotation. The image spatial resolution varies from 1 meter to 10 meters depending on the ordered product. A TerraSAR-X image has to be ordered in a selectable data representation, where four main alternative representations are available (Single look Slant range Complex (SSC), Multi-look Ground range Detected (MGD), Geo-coded Ellipsoid Corrected (GEC) and Enhanced Ellipsoid Corrected (EEC))
To investigate the capability of SAR data for extensive urban area and slum mapping, this study uses two scenes of partially polarized SAR data acquired with the TerraSAR-X (TSX) and TanDEM-X (TDX) satellites. The study area is located in the Indian megacity Mumbai which is home to millions of slum dwellers. In order to train and validate the classification models an area- wide reference data set is created. The informal settlements were delineated in a previous study by visual interpretation of very highresolution optical images (Taubenböck and Wurm 2015) and adapted to the acquisition date of the SAR imagery in this study. Furthermore, urban areas are derived from the Global Urban Footprint (GUF), a binary mask holding all built-up man-made structures (Esch et al. 2013). A third class representing all other land cover, e.g. water, vegetation or bare soil, is assigned to all remaining areas. This reference is used for the training of two different classifiers, Random Forest (RF) and Linear Discriminant Analysis (LDA).