Synthetic Aperture Radar Datasets Research Articles

Hybrid Variability Aware Network (HVANet): A Self-Supervised Deep Framework for Label-Free SAR Image Change Detection

Synthetic aperture radar (SAR) image change detection (CD) aims to automatically recognize changes over the same geographic region by comparing prechange and postchange SAR images. However, the detection performance is usually subject to several restrictions and problems, including the absence of labeled SAR samples, inherent multiplicative speckle noise, and class imbalance. More importantly, for bitemporal SAR images, changed regions tend to present highly variable sizes, irregular shapes, and different textures, typically referred to as hybrid variabilities, further bringing great difficulties to CD. In this paper, we argue that these internal hybrid variabilities can also be used for learning stronger feature representation, and we propose a hybrid variability aware network (HVANet) for completely unsupervised label-free SAR image CD by taking inspiration from recent developments in deep self-supervised learning. First, since different changed regions may exhibit hybrid variabilities, it is necessary to enrich distinguishable information within the input features. To this end, in shallow feature extraction, we generalize the traditional spatial patch (SP) feature to allow for each pixel in bitemporal images to be represented at diverse scales and resolutions, called extended SP (ESP). Second, with the carefully customized ESP features, HVANet performs local spatial structure information extraction and multiscale–multiresolution (MS-MR) information encoding simultaneously through a local spatial stream and a scale-resolution stream, respectively. Intrinsically, HVANet projects the ESP features into a new high-level feature space, where the change identification becomes easier. Third, to train the framework effectively, a self-supervision layer is attached to the top of the HVANet to enable the two-stream feature learning and recognition of changed pixels in the corresponding feature space, in a self-supervised manner. Experimental results on three low/medium-resolution SAR datasets demonstrate the effectiveness and superiority of the proposed framework in unsupervised SAR CD tasks.

Addressing critical influences on L-band radar backscatter for improved estimates of basal area and change

L-band synthetic aperture radar (SAR) backscatter intensity is sensitive to land cover and can be used to estimate vegetation measures such as basal area (BA) and biomass. However, the estimation of BA, and especially change in BA, can be hampered by the influences upon backscatter of external factors such as imaging geometry, terrain topology, prevailing moisture conditions and even SAR sensor characteristics. This paper describes a method of reducing the adverse effects of such extraneous influences on vegetation and change estimates derived from single-channel SAR data. Empirical corrections for terrain slope and cross-track tendencies were applied and linear least squares difference minimization used to normalize the backscatter differences between scenes. The method was applied to state-wide coverage of L-band, fine-mode, HV polarization Advanced Land Observing Satellite (ALOS) Phased Array L-band SAR (PALSAR) data over New South Wales (NSW), Australia. The data were acquired with different sensors over two “observational epochs”: ALOS PALSAR in 2009 and ALOS-2 PALSAR-2 in 2016/17. The SAR datasets presented significant variations in backscatter intensity beyond those attributable to changes in vegetation cover. The corrective procedures resulted in improved uniformity of observed backscatter dependence on vegetation. Variations in backscattering coefficient between swaths were reduced by as much as 1.75 dB and 25% of the standard deviation in mean backscattering coefficients in common areas and at near- and far-range. This corresponded to a correction in BA estimate of 4.4 m2 ha−1. The method was observed to reduce ambiguities in regrowth estimates at swath boundaries and correct estimates of BA change by as much as 30% over large areas. The resulting estimates of 7-year change in BA provide spatially explicit forest structural information that is assisting in monitoring changes in woody vegetation across NSW.

Farmland parcel-based crop classification in cloudy/rainy mountains using Sentinel-1 and Sentinel-2 based deep learning

ABSTRACT Multitemporal remote sensing data, especially those for key phenological periods, play an important role in crop classification. However, cloudy/rainy climate conditions can easily lead to a lack of valid optical data, leading to crop classification difficulties. A general solution is taking advantage of all-weather synthetic aperture radar (SAR) datasets. In practice, SAR and optical datasets are often applied in the agricultural field by the method of image fusion, but it is difficult to apply when the number of optical images is too small. To solve this problem, this research proposes a data-transfer and feature-optimize-based method, which deploy an RNN-based encoding-decoding network to add additional data to the ‘optical’ temporal features at the farmland parcel scale and improve the utilization of optical fragments. On the basis of this method, we mitigate inconsistencies in spatial scale among different datasets and optimize the time-series parameters without expert knowledge in the crop classification procedure. The experimental results illustrate the crop classification accuracy of this method, which achieves a 4.1% improvement over the traditional approach and is especially effective for dryland crops (e.g. corn and rapeseed). Thus, this research demonstrates the effectiveness of the combined use of optical and SAR data for similar applications in cloudy/rainy mountainous areas.

The comparative study of three nonparametric methods of SAR tomography for building reconstruction

Over the past years, synthetic aperture radar (SAR) tomography (TomoSAR) has indicated significant potentials for three-dimensional (3D) reconstruction of buildings in urban areas. A large number of SAR images are thus required to perform tomographic inversions in non-parametric spectral estimators. To this end, in the case of a limited choice of SAR acquisitions, the present study aims to evaluate the capabilities of five non-parametric spectral estimation (SE) techniques, including linear prediction (LP), maximum entropy (ME), and minimum norm (MN), Capon, and beamforming (BF) in the tomographic reconstruction of urban environments. The performance analysis is carried out by using both simulated and real SAR datasets. The study results demonstrate that the proposed efficient LP estimator, minimizing the average output signal power over the antenna array elements, can separate the layover scatterers along the height direction. This low-computational SE method is thus able to clean side lobes while using a small number of observations. The proposed algorithms, as applied on TerraSAR-X strip-map images of the city of Tehran, Iran, verify the effectiveness of the non-parametric LP reconstruction technique for urban buildings. The estimated height of the scatterers also indicates that the LP estimator is similar to the field-based measurements once compared with ME, MN, Capon, and BF reconstruction methods.

초해상화 기반 CNN을 이용한 군사용 SAR 자동표적인식 모델 연구

Herein, we propose employing the super-resolution-based convolutional neural network (CNN) to design the automatic target recognition (ATR) of military synthetic aperture radar (SAR) images. Previous SAR ATR methods showed a good recognition performance with a low depression angle, but poor performance with a high depression angle. In a warfighting environment, good recognition performance is required even with a high depression angle. To address this issue, we combine the super-resolution method and the CNN. In comparison with the conventional VGGnet with a high depression angle, the proposed super-resolution-based CNN showed a 3%-4% improvement in accuracy. The MSTAR SAR dataset was utilized for validation.

Spatio-temporal evolution of the magma plumbing system at Masaya Caldera, Nicaragua

Volcanic unrest in calderas can be exhibited through a variety of different mechanisms, such as changes in seismicity and ground deformation, as well as variations in thermal and/or gas emissions. However, not all caldera unrest results in explosive caldera-forming volcanic activity. Alternative activity may include periods of quiescence, passive degassing, effusive activity (e.g., lava flows lava lakes and dome formation), and/or magma injection into the shallow magma system. In this study, we perform a long-term study (spanning 2011–2019) of ground deformation at Masaya using six Interferometric Synthetic Aperture Radar (InSAR) datasets. Masaya exhibited bi-modal eruptive behavior between 2011 and 2019, dominated by open-vent lava lake activity and punctuated by short-lived summit explosions. The Multidimensional Small BAseline Subset time-series analysis approach was used to take advantage of the temporally dense SAR datasets. Between 2012 and early 2015, we observed degassing-induced pressurization of the Masaya Central Reservoir (MCR) at an estimated volume change rate of ~ 0.28 $$\times$$ 106 m3/year. In May 2015, magma was supplied into the MCR at a rate of ~ 5.6 $$\times$$ 106 m3/year, leading to the appearance of a summit lava lake in December 2015. Over the next 6 months, rapid magma supply continued to drive lava lake activity and was followed by a cessation of magma supply into the MCR for another 11 months. From mid-2017 to end-2019, we observed depressurization (~ − 0.67 $$\times$$ 106 m3/year) of the MCR due to a lack of magma supply and continued high rates of degassing in-conjunction with declining lava lake activity.

Time-Series Clustering Methodology for Estimating Atmospheric Phase Screen in Ground-Based InSAR Data

In multitemporal interferometric synthetic aperture radar (InSAR) applications, propagation delay in the troposphere introduces a major source of disturbance known as atmospheric phase screen (APS). This study proposes a novel framework to compensate for the APS from multitemporal ground-based InSAR data. The proposed framework first performs time-series clustering in accordance with the temporal APS behavior realized by the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$k$ </tex-math></inline-formula> -means clustering approach. In the second step, joint estimation of the APS and displacement velocity is performed. For this purpose, a novel interferometric signal model, including the APS modeled by the median profiles defined in each cluster, is proposed. The proposed framework is validated with the Ku-band ground-based synthetic aperture radar data sets measured over a mountainous area in Kumamoto, Japan. Tests on these data sets reveal that compared with the conventional approach, the presented approach improves displacement estimation accuracy under severe atmospheric conditions.

Understanding the Slow Motion of the Wangjiashan Landslide in the Baihetan Reservoir Area (China) from Space‐Borne Radar Observations

The analysis of landslide evolution using archived optical remote‐sensing images is common, but it is often limited by the acquisition frequency, cloud cover, and resolution. With the development of space‐borne radar observation technology, small baseline subset interferometric synthetic aperture radar (SAR) technology provides a new technical approach for detecting landslide deformation before disasters. The Sentinel‐1A SAR datasets (20170219–20210330) were used to study the time‐series deformation characteristics of the Wangjiashan landslide in the Baihetan Reservoir area before its impoundment. The time‐series results show that the Wangjiashan landslide was in an initial deformation state in the prior four years, and the deformation first occurred in the middle part and then expanded to the landslide toe and crown retrogressive movement characteristics. Combined with an analysis of field deformation signs, these findings suggest that the upper landslide mass formed a local sliding surface, which caused serious deformation of the road. An analysis of historical rainfall data revealed that the Wangjiashan landslide is sensitive to rainfall, and the deformation is not only significantly correlated with cumulative rainfall but also influenced by concentrated heavy rainfall. The research in this study provides a basis for the monitoring, early warning, and risk management of the Wangjiashan landslide during the impoundment period. This work also provides a useful reference for investigations using space‐borne SAR Earth observations in geological disaster prevention and control.

Change Detection From Synthetic Aperture Radar Images via Graph-Based Knowledge Supplement Network

Synthetic aperture radar (SAR) image change detection is a vital yet challenging task in the field of remote sensing image analysis. Most previous works adopt a self-supervised method which uses pseudolabeled samples to guide subsequent training and testing. However, deep networks commonly require many high-quality samples for parameter optimization. The noise in pseudolabels inevitably affects the final change detection performance. To solve the problem, we propose a graph-based knowledge supplement network (GKSNet). To be more specific, we extract discriminative information from the existing labeled dataset as additional knowledge, to suppress the adverse effects of noisy samples to some extent. Afterward, we design a graph transfer module to distill contextual information attentively from the labeled dataset to the target dataset, which bridges feature correlation between datasets. To validate the proposed method, we conducted extensive experiments on four SAR datasets, which demonstrated the superiority of the proposed GKSNet as compared to several state-of-the-art baselines. Our codes are available: at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/summitgao/SAR_CD_GKSNet</uri> .

Multimodal Semantic Consistency-Based Fusion Architecture Search for Land Cover Classification

Multimodal Land Cover Classification (MLCC) using the optical and Synthetic Aperture Radar (SAR) modalities has resulted in outstanding performances over using only unimodal data due to their complementary information on land properties. Previous multimodal deep learning (MDL) methods have relied on handcrafted multi-branch convolutional neural networks (CNN) to extract the features of different modalities and merged them for land cover classification. However, natural images-oriented handcrafted CNN models may not the optimal strategies to handle Remote Sensing (RS) image interpretation problems, due to the huge difference in terms of imaging angles and imaging ways. Furthermore, few MDL methods have analyzed optimal combinations of hierarchical features from different modalities. In this article, we propose an efficient multimodal architecture search framework, namely Multimodal Semantic Consistency-Based Fusion Architecture Search (M²SC-FAS) in continuous search space with the gradient-based optimization method, which can not only discover optimal optical- and SAR-specific architectures according to the different characteristics of the optical and SAR images, respectively, but also realizes the search of optimal multimodal dense fusion architecture. Specifically, the semantic-consistency constraint is introduced to guarantee dense fusion between hierarchical optical and SAR features with high semantic consistency and then capture the complementary performance on land properties. Finally, the basis of curriculum learning strategy is adopted on the M²SC-FAS. Extensive experiments show superior performances of our work on three broad co-registered optical and SAR datasets.

Robust Rayleigh Regression Method for SAR Image Processing in Presence of Outliers

The presence of outliers (anomalous values) in synthetic aperture radar (SAR) data and the misspecification in statistical image models may result in inaccurate inferences. To avoid such issues, the Rayleigh regression model based on a robust estimation process is proposed as a more realistic approach to model this type of data. This article aims at obtaining Rayleigh regression model parameter estimators robust to the presence of outliers. The proposed approach considered the weighted maximum likelihood method and was submitted to numerical experiments using simulated and measured SAR images. Monte Carlo simulations were employed for the numerical assessment of the proposed robust estimator performance in finite signal lengths, their sensitivity to outliers, and the breakdown point. For instance, the nonrobust estimators show a relative bias value 65-fold larger than the results provided by the robust approach in corrupted signals. In terms of sensitivity analysis and break down point, the robust scheme resulted in a reduction of about 96% and 10%, respectively, in the mean absolute value of both measures, in compassion to the nonrobust estimators. Moreover, two SAR datasets were used to compare the ground type and anomaly detection results of the proposed robust scheme with competing methods in the literature.

Explore Better Network Framework for High-Resolution Optical and SAR Image Matching

To fully explore the complementary information from optical and synthetic aperture radar (SAR) imageries, they need first to be coregistered with high accuracy. Due to the vast radiometric and geometric disparity, the problem to match high-resolution optical and SAR images is quite challenging. The present deep learning-based methods have shown advantages over the traditional approaches, but the performance increment is not significant. In this article, we explore a better network framework for high-resolution optical and SAR image matching from three aspects. First, we propose an effective multilevel feature fusion method, which helps to take advantage of both the low-level fine-grained features for precious feature location and the high-level semantic features for better discriminative ability. Second, a feature channel excitation procedure is conducted using a novel multifrequency channel attention module, which is able to make image features of different types and multiple levels effectively collaborate with each other and produce image matching features with high diversity. Third, the self-adaptive weighting loss is introduced, with which, each sample is assigned with an adaptive weighting factor, and therefore, information buried in all nearby samples can be better exploited. Under a pseudo-Siamese architecture, the proposed optical and SAR image matching network (OSMNet) is trained and tested on a large and diverse high-resolution optical and SAR dataset. Extensive experiments demonstrate that each component of the proposed deep framework helps to improve the matching accuracy. Also, the OSMNet shows overwhelming superior to the state-of-the-art handcrafted approaches on imageries of different land-cover types.

An Entropy-Weighted Network for Polar Sea Ice Open Lead Detection From Sentinel-1 SAR Images

Sea ice leads in the Arctic Ocean and Antarctic Ocean are of great significance to polar ecology, climate change, and ship navigation. While rule-based remote-sensing classification methods, such as the threshold method, have been widely used for surveying and mapping polar sea ice leads, they have difficulty in overcoming the problems of noise and poor generalization. In this article, we presented an automated, deep learning sea ice open lead detection network in low wind speed conditions, entropy-weighted network (EW-Net). In EW-Net, a dense block was introduced into a U-Net baseline network to strengthen feature propagation, and an entropy sampling structure was designed to improve the pertinence of the network to leads and alleviate the influence of noise. Furthermore, an entropy-weighted feature fusion block was designed to better restore features. EW-Net was trained on the Sentinel-1 synthetic aperture radar (SAR) dataset in order to obtain lead maps of different polar regions from Sentinel-1 images. The results showed that the proposed EW-Net model was more effective and more generalizable for polar sea ice lead detection than the threshold method and the deep learning methods U-Net and DeepLab V3 Plus, with an overall accuracy (OA) exceeding 0.97. In addition, EW-Net showed advantages in terms of outstanding generalization from cross-sensor image monitoring of leads. The method proposed in this article could be beneficial for providing accurate maps of polar sea ice leads in open water and contribute to further research on polar science.

Hierarchical Dictionary Learning for Vehicle Classification Based on the Carrier-Free UWB Radar

As a promising technique, dictionary learning (DL) for target recognition has seen a recent surge in recent years. Although many methods have been proposed to obtain discriminative dictionaries or coefficients via incorporating various constraints into the objective function, there are still two issues. First, it is well known that kinds of discriminative criteria on the objective function often involve substantial optimized items, increasing computation cost. Second, noises in the real world inevitably degrade the classification performance, while most DL algorithms disregard that. Aiming at these two problems, a hierarchical DL model is proposed for vehicle recognition based on the carrier-free ultrawideband (UWB) radar. With the purpose of successfully determining the identity of targets, we first learn several class-specific subdictionaries. Then, considering that the actual environment is filled with noises, we divided the learned dictionary atoms into signal and disturbance atoms in accordance with sparse coefficients to establish the signal dictionary and noise dictionary, respectively. Finally, the clean data are recovered over the corresponding signal dictionary, and meanwhile, the classification task is achieved. This hierarchical DL method takes into account both the noise-robust ability and discriminative power of the learned dictionary, in which the “atom selection” mechanism dramatically speeds up calculations. What is more, rather than imposing discriminative restraints on the objective function, we improve the K-SVD-based optimization process to complete hierarchical DL. Experimental results on the measured and synthetic data corroborate the effectiveness of the proposed method even under low signal-to-noise ratio (SNR) values. Especially, to testify to the generalization ability of the proposed method, we evaluate our algorithm on a public synthetic aperture radar (SAR) dataset (MSTAR).

WetNet: A Spatial–Temporal Ensemble Deep Learning Model for Wetland Classification Using Sentinel-1 and Sentinel-2

While deep learning models have been extensively applied to land-use land-cover (LULC) problems, it is still a relatively new and emerging topic for separating and classifying wetland types. On the other hand, ensemble learning has demonstrated promising results in improving and boosting classification accuracy. Accordingly, this study aims to develop a classification system for mapping complex wetland areas by incorporating deep ensemble learning and satellite datasets. To this end, time series of Sentinel-1 dual-polarized Synthetic Aperture Radar (SAR) dataset, alongside Sentinel-2 multispectral imagery (MSI), are used as input data to the model. In order to increase the diversity of the extracted features, the proposed model, herein called WetNet, consists of three different submodels, comprising several recurrent and convolutional layers. Furthermore, multiple ensembling sections are added to different stages of the model to increase the transferability of the model (to other areas) and the reliability of the final results. WetNet is evaluated in a complex wetland area located in Newfoundland, Canada. Experimental results indicate that WetNet outperforms the state-of-the-art deep models (e.g., InceptionResnetV2, InceptionV3, and DenseNet121) in terms of both the classification accuracy and processing time. This makes WetNet an efficient model for large-scale wetland mapping application. The python code of the proposed WetNet model is available at the following link for the sake of reproducibility: <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://colab.research.google.com/drive/1pvMOd3_tFYaMYGyHNfxqDxOiwF78lKgN?usp=sharing</uri>

A Multigrid InSAR Technique for Joint Analyses at Single-Look and Multi-Look Scales

This work proposes a multigrid differential synthetic aperture radar (SAR) interferometry (InSAR) technique for the detection of ground displacements at different spatial scales. The method relies on efficient phase-unwrapping (PhU) operations performed at the native spatial scales. In particular, a set of multi-look interferograms are first unwrapped using conventional (or advanced) PhU algorithms at the regional scale. Subsequently, ML unwrapped interferograms are used to facilitate the PhU operations performed at the local scale (single-look). Specifically, the unwrapped multi-look interferograms are resampled to the single-look grid and modulo- <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$2\pi $ </tex-math></inline-formula> subtracted to the single-look interferograms. These phase residuals are then unwrapped and added back to the multi-look resampled interferograms. To accomplish these operations, at variance with alternative multiscale methods, no (linear/nonlinear) models are used to fit the spatial high-pass phase residuals. Finally, the unwrapped single-look interferograms are properly inverted to retrieve the ground displacement time series using any small baseline (SB)-oriented multitemporal InSAR tool. Experimental results are performed by processing a set of SAR data acquired by the X-band COSMO-SkyMed sensor over the coastal area of Shanghai, China.

Improved Point Estimation for the Rayleigh Regression Model

The Rayleigh regression model was recently proposed for modeling amplitude values of synthetic aperture radar (SAR) image pixels. However, inferences from such model are based on the maximum likelihood estimators, which can be biased for small signal lengths. The Rayleigh regression model for SAR images often takes into account small pixel windows, which may lead to inaccurate results. In this letter, we introduce bias-adjusted estimators tailored for the Rayleigh regression model based on: (i) the Cox and Snell's method; (ii) the Firth's scheme; and (iii) the parametric bootstrap method. We present numerical experiments considering synthetic and actual SAR data sets. The bias-adjusted estimators yield nearly unbiased estimates and accurate modeling results.

Spatial–Temporal Gray-Level Co-Occurrence Aware CNN for SAR Image Change Detection

Deep learning-based synthetic aperture radar (SAR) image change detection has recently achieved remarkable success due to its great potential for extracting abstract features. However, the existing methods still have room for improvement in dealing with the speckle of SAR images. In this letter, a deep spatial–temporal gray-level co-occurrence aware convolutional neural network (STGCNet) is proposed, which can effectively mine the spatial–temporal information of the bitemporal images and obtain the speckle-robust results by introducing the 3-D gray-level co-occurrence matrix (3-D-GLCM) as auxiliary feature. Specifically, representative features are extracted from original image pairs and their corresponding 3-D-GLCM through two-stream network, followed by an adaptive fusion module to balance the contribution of each branch. Then, the final binary change detection results are obtained by a fully connected layer. The training process relies on reliable labels generated by unsupervised models rather than manually annotated data, and therefore, the proposed STGCNet is practical in reality. Experiments on synthesized and real SAR data sets demonstrate the robustness and competitiveness of the proposed method compared with the state-of-the-art algorithms.

Change Detection in Synthetic Aperture Radar Images Using a Dual-Domain Network

Change detection from synthetic aperture radar (SAR) imagery is a critical yet challenging task. Existing methods mainly focus on feature extraction in the spatial domain, and little attention has been paid to the frequency domain. Furthermore, in patch-wise feature analysis, some noisy features in the marginal region may be introduced. To tackle the above two challenges, we propose a dual-domain network (DDNet). Specifically, we take features from the discrete cosine transform (DCT) domain into consideration and the reshaped DCT coefficients are integrated into the proposed model as the frequency domain branch. Feature representations from both frequency and spatial domain are exploited to alleviate the speckle noise. In addition, we further propose a multi-region convolution (MRC) module, which emphasizes the central region of each patch. The contextual information and central region features are modeled adaptively. The experimental results on three SAR data sets demonstrate the effectiveness of the proposed model. Our codes are available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/summitgao/SAR_CD_DDNet</uri> .

SAR Image Change Detection Based on Heterogeneous Graph With Multiattributes and Multirelationships

The performance of change detection between synthetic aperture radar (SAR) images mainly depends on the selection and utilization of image attributes. Nevertheless, most existing change detection approaches merely take the intensity attribute into consideration, constraining their capacities of detecting changes in complex situations. To solve this problem, this study develops an unsupervised SAR image change detection approach based on heterogeneous graph with multi-attributes and multi-relationships. First, the structural attribute, obtained by removing the complex texture, can be used to reflect the overall features of a SAR image. A heterogeneous graph is then constructed to encode the structural and intensity attributes as vertices, and two types of edges are hence connected to capture the intra-relationships and inter-relationships from these different attributes. With the support of this graph, the hyper-adjacency characteristics are proposed to quantify the multi-relationships and describe the global heterogeneous information. Constructing heterogeneous graphs respectively on bi-temporal SAR images, the changes of both intensity and structural attributes can be measured via comparing the hyper-adjacency characteristics of the bi-temporal graphs, thereby generating a difference image with good separability. Finally, the change map is obtained by cutting off the weakly related edges of heterogeneous graph on the difference image. Experiments on four real SAR datasets prove the effectiveness of the proposed graph-driven approach in improving the accuracy and robustness of change detection.