Synthetic Aperture Radar Datasets Research Articles

Classifying Ships in SAR Images by Using Contour Bias Features and Transfer Learning

With the development of synthetic-aperture radar (SAR) image interpretation technology in recent years, classifying detected ships based on SAR images has become an important trend in ocean monitoring. However, owing to the underlying imaging mechanism, the texture of SAR images typically contains noise that cannot be easily eliminated. Therefore, in this study, we consider more stable target contour features to classify SAR images of marine vessels. Based on a deep learning algorithm, we propose a method to obtain contour bias features by employing style transfer learning to classify detected ships from SAR data. We also adopt transfer learning to improve the uneven distribution of SAR datasets of representative ships. The results of experiments conducted to evaluate the proposed approach show that contour bias features improve the generalization performance and classification accuracy of the model. They also show that transfer learning effectively avoids the problem of data imbalance and, thus, improves classification accuracy on the OpenSARShip 2.0 database.

Few Samples of SAR Automatic Target Recognition Based on Enhanced-Shape CNN

Synthetic Aperture Radar (SAR), as one of the important and significant methods for obtaining target characteristics in the field of remote sensing, has been applied to many fields including intelligence search, topographic surveying, mapping, and geological survey. In SAR field, the SAR automatic target recognition (SAR ATR) is a significant issue. However, on the other hand, it also has high application value. The development of deep learning has enabled it to be applied to SAR ATR. Some researchers point out that existing convolutional neural network (CNN) paid more attention to texture information, which is often not as good as shape information. Wherefore, this study designs the enhanced-shape CNN, which enhances the target shape at the input. Further, it uses an improved attention module, so that the network can highlight target shape in SAR images. Aiming at the problem of the small scale of the existing SAR data set, a small sample experiment is conducted. Enhanced-shape CNN achieved a recognition rate of 99.29% when trained on the full training set, while it is 89.93% on the one-eighth training data set.

The Global Water Body Layer from TanDEM-X Interferometric SAR Data

The interferometric synthetic aperture radar (InSAR) data set, acquired by the TanDEM-X (TerraSAR-X add-on for Digital Elevation Measurement) mission (TDM), represents a unique data source to derive geo-information products at a global scale. The complete Earth’s landmasses have been surveyed at least twice during the mission bistatic operation, which started at the end of 2010. Examples of the delivered global products are the TanDEM-X digital elevation model (DEM) (at a final independent posting of 12 m × 12 m) or the TanDEM-X global Forest/Non-Forest (FNF) map. The need for a reliable water product from TanDEM-X data was dictated by the limited accuracy and difficulty of use of the TDX Water Indication Mask (WAM), delivered as by-product of the global DEM, which jeopardizes its use for scientific applications, as well. Similarly as it has been done for the generation of the FNF map; in this work, we utilize the global data set of TanDEM-X quicklook images at 50 m × 50 m resolution, acquired between 2011 and 2016, to derive a new global water body layer (WBL), covering a range from −60∘ to +90∘ latitudes. The bistatic interferometric coherence is used as the primary input feature for performing water detection. We classify water surfaces in single TanDEM-X images, by considering the system’s geometric configuration and exploiting a watershed-based segmentation algorithm. Subsequently, single overlapping acquisitions are mosaicked together in a two-step logically weighting process to derive the global TDM WBL product, which comprises a binary averaged water/non-water layer as well as a permanent/temporary water indication layer. The accuracy of the new TDM WBL has been assessed over Europe, through a comparison with the Copernicus water and wetness layer, provided by the European Space Agency (ESA), at a 20 m × 20 m resolution. The F-score ranges from 83%, when considering all geocells (of 1∘ latitudes × 1∘ longitudes) over Europe, up to 93%, when considering only the geocells with a water content higher than 1%. At global scale, the quality of the product has been evaluated, by intercomparison, with other existing global water maps, resulting in an overall agreement that often exceeds 85% (F-score) when the content in the geocell is higher than 1%. The global TDM WBL presented in this study will be made available to the scientific community for free download and usage.

Accuracy of Sentinel-1 PSI and SBAS InSAR Displacement Velocities against GNSS and Geodetic Leveling Monitoring Data

Correct use of multi-temporal Interferometric Synthetic Aperture Radar (InSAR) datasets to complement geodetic surveying for geo-hazard applications requires rigorous assessment of their precision and accuracy. Published inter-comparisons are mostly limited to ground displacement estimates obtained from different algorithms belonging to the same family of InSAR approaches, either Persistent Scatterer Interferometry (PSI) or Small BAseline Subset (SBAS); and accuracy assessments are mainly focused on vertical displacements or based on few Global Navigation Satellite System (GNSS) or geodetic leveling points. To fill this demonstration gap, two years of Sentinel-1 SAR ascending and descending mode data are processed with both PSI and SBAS consolidated algorithms to extract vertical and horizontal displacement velocity datasets, whose accuracy is then assessed against a wealth of contextual geodetic data. These include permanent GNSS records, static GNSS benchmark repositioning, and geodetic leveling monitoring data that the National Institute of Statistics, Geography, and Informatics (INEGI) of Mexico collected in 2014−2016 in the Aguascalientes Valley, where structurally-controlled land subsidence exhibits fast vertical rates (up to −150 mm/year) and a non-negligible east-west component (up to ±30 mm/year). Despite the temporal constraint of the data selected, the PSI-SBAS inter-comparison reveals standard deviation of 6 mm/year and 4 mm/year for the vertical and east-west rate differences, respectively, thus reassuring about the similarity between the two types of InSAR outputs. Accuracy assessment shows that the standard deviations in vertical velocity differences are 9−10 mm/year against GNSS benchmarks, and 8 mm/year against leveling data. Relative errors are below 20% for any locations subsiding faster than −15 mm/year. Differences in east-west velocity estimates against GNSS are on average −0.1 mm/year for PSI and +0.2 mm/year for SBAS, with standard deviations of 8 mm/year. When discrepancies are found between InSAR and geodetic data, these mostly occur at benchmarks located in proximity to the main normal faults, thus falling within the same SBAS ground pixel or closer to the same PSI target, regardless of whether they are in the footwall or hanging wall of the fault. Establishing new benchmarks at higher distances from the fault traces or exploiting higher resolution SAR scenes and/or InSAR datasets may improve the detection of the benchmarks and thus consolidate the statistics of the InSAR accuracy assessments.

New insights on faulting and intrusion processes during the June 2007, East Rift Zone eruption of Kīlauea volcano, Hawai'i

The East Rift Zone (ERZ) of Kīlauea Volcano, Hawai'i, represents one of the most volcanically active regions in the world. The 2007 Father's Day (FD) dike intrusion, eruption, and accompanying slow-slip event (SSE) has been previously modeled using geodetic data to constrain the geometry of the intrusion and the timing and magnitude of the SSE. Here, we perform inversions of three interferometric synthetic aperture radar (InSAR) datasets and a new intensity offset tracking dataset to assess the effect of integrating intensity cross-correlation offsets into inversion problems and explore additional potential models for the intrusion geometry of the FD event based on this additional data. The overall lowest misfit single Okada model for all datasets opens 2.3 m, strikes 73 degrees while dipping sub-vertically at 83 degrees, and extends approximately 2.9 km to the ENE and 2.4 km downdip. The differences are minor between complex en-echelon distributed Okada and decollement model of (Montgomery-Brown et al., 2010) or 3D-MBEM breaching models including multiple surface breaches and free-slipping decollement movement. Finally, we examine the static Coulomb stress changes for the proposed decollement fault created by our preferred model and a representative model of deep rift opening and find that deep rift zones dilation, not shallow ERZ intrusions, are likely modulating slip on the decollement.

Three-dimensional and long-term landslide displacement estimation by fusing C- and L-band SAR observations: A case study in Gongjue County, Tibet, China

Recently, a large number of synthetic aperture radar (SAR) images has been introduced into landslide investigations with the growing launch of new SAR satellites, such as ALOS/PALSAR-2 and Sentinel-1. Therefore, it is appropriate to develop new approaches to retrieve three-dimensional (3D) displacements and long-term (> 10 years) displacement time series to investigate the spatio-temporal evolution and creep behavior of landslides. In this study, a new approach for the estimation of 3D and long-term displacement time series of landslides, based on the fusion of C- and L-band SAR observations, is presented. This method is applied to map 3D and long-term displacements (nearly 12 years) of the landslides in Gongjue County, Tibet in China; four sets of SAR images from different platforms (i.e., L-band ascending ALOS/PALSAR-1, C-band descending ENVISAT, and C-band ascending and descending Sentinel-1 SAR datasets) covering the period of January 2007 to November 2018 were collected and exploited. First, the assumption that the landslide moves parallel to its ground surface is used to produce 3D displacement rates and time series by fusing ascending and descending Sentinel-1 SAR images, from which the optimal sliding direction for each pixel of the slope is well estimated. Then, the long-term displacement time-series of the landslide between January 2007 and October 2018 in the estimated sliding direction is recovered by fusing L-band ALOS/PALSAR-1 and C-band Sentinel-1 SAR images. In order to fill the time gap of nearly four years between ALOS/PALSAR-1 and Sentinel-1 SAR images, the Tikhonov regularization (TR) method is developed to establish the observational equation. Moreover, to solve the problem arising from ALOS/PALSAR-1 and Sentinel-1 images with different wavelengths, incidence angles and flight directions, the measurements from ALOS/PALSAR-1 and Sentinel-1 images are both projected to the estimated optimal sliding direction to achieve a unified displacement datum. Our results from ascending and descending Sentinel-1 images suggest that the maximum displacement rates of the study area in the vertical and east-west directions from December 2016 to October 2018 were greater than 70 and 80 mm/year, respectively, and 2D displacement results reveal that the displacement patterns and movement characteristics of all the detected landslides are not identical in the study area. Specifically, the 3D displacement results successfully revealed the spatiotemporal displacement patterns and movement direction of each block of the Shadong landslide, and long-term displacement time series showed for the first time that the maximum cumulative displacement exceeds 1.3 m from January 2007 to October 2018. Moreover, the kinematic evolution and possible driving factors of landslides were investigated using 2D and 3D and long-term displacement results, coupled with hydrological factors and unidimensional constitutive models of the rocks.

Spatial distribution of inter- and intra-crop variability using time-weighted dynamic time warping analysis from Sentinel-1 datasets

This study investigates the robustness of an advanced classification algorithm to spatially map heterogeneous, fragmented croplands using multi-temporal synthetic aperture radar (SAR) datasets. Four parameters derived from Sentinel-1 (backscatters in dual-polarization: σVHo, σVVo, cross-ratio (σVHoσVVo), and radar vegetation index (RVI)) were considered to develop temporal patterns and correlate with un-classified time-series satellite imagery using time-weighted dynamic time warping (TWDTW) algorithm. Pixel and parcel based classifications were considered to identify four crop varieties (paddy, sugarcane, cotton and vegetables) subjected to two water limiting conditions (low stress-LS, high stress-HS). In-situ data were split-sampled (30:70 ratio) between training (to develop temporal patterns) and testing (to validate classification output). Overall accuracy of pixel and parcel based classifications were 63% and 76% with a Kappa coefficient of 0.58 and 0.73 respectively. In conclusion, parcel based TWDTW algorithm conditioned by temporal signatures of RVI has effectively delineated croplands with varying irrigation treatments for yield and damage assessment modeling studies.

Semisupervised SAR image change detection based on a siamese variational autoencoder

In synthetic aperture radar (SAR) image change detection, the deep learning has attracted increasingly more attention because the difference images (DIs) of traditional unsupervised technology are vulnerable to speckle noise. However, most of the existing deep networks do not constrain the distributional characteristics of the hidden space, which may affect the feature representation performance. This paper proposes a variational autoencoder (VAE) network with the siamese structure to detect changes in SAR images. The VAE encodes the input as a probability distribution in the hidden space to obtain regular hidden layer features with a good representation ability. Furthermore, subnetworks with the same parameters and structure can extract the spatial consistency features of the original image, which is conducive to the subsequent classification. The proposed method includes three main steps. First, the training samples are selected based on the false labels generated by a clustering algorithm. Then, we train the proposed model with the semisupervised learning strategy, including unsupervised feature learning and supervised network fine-tuning. Finally, input the original data instead of the DIs in the trained network to obtain the change detection results. The experimental results on four real SAR datasets show the effectiveness and robustness of the proposed method.

Combined GRACE and MT-InSAR to Assess the Relationship between Groundwater Storage Change and Land Subsidence in the Beijing-Tianjin-Hebei Region

Beijing-Tianjin-Hebei (BTH) has been suffering from severe groundwater storage (GWS) consumption and land subsidence (LS) for a long period. The overexploitation of groundwater brings about severe land subsidence, which affects the safety and development of BTH. In this paper, we utilized multi-frame synthetic aperture radar datasets obtained by the Rardarsat-2 satellite to monitor land subsidence’s temporal and spatial distribution in the BTH from 2012 to 2016 based on multi-temporal interferometric synthetic aperture radar (MT-InSAR). In addition, we also employed the Gravity Recovery and Climate Experiment (GRACE) mascon datasets acquired by the Center for Space Research (CSR) and Jet Propulsion Laboratory (JPL) to obtain the GWS anomalies (GWSA) of BTH from 2003 to 2016. Then we evaluate the accuracy of the results obtained. Furthermore, we explored the relationship between the regional GWSA and the average cumulative subsidence in the BTH. The total volume change of subsidence is 59.46% of the total volume change of groundwater storage. Moreover, the long-term decreasing trend of the GWSA (14.221 mm/year) and average cumulative subsidence (17.382 mm/year) show a relatively high consistency. Finally, we analyze the heterogeneity of GWS change (GWSC) and LS change (LSC) in the four typical areas by the Lorenz curve model. The implementation of the South-to-North Water Diversion Project (MSWDP) affects the heterogeneity of GWSC and LSC. It can be seen that the largest heterogeneity of LSC lags behind the GWSC in the Tianjin-Langfang-Hengshui-Baoding area. The largest uneven subsidence in Beijing and Tianjin occurred in 2015, and the largest uneven subsidence in Hengshui-Baoding occurred in 2014. After that, the heterogeneity of subsidence gradually tends to stable.

Change Detection from SAR Images Based on Convolutional Neural Networks Guided by Saliency Enhancement

Change detection is an important task in identifying land cover change in different periods. In synthetic aperture radar (SAR) images, the inherent speckle noise leads to false changed points, and this affects the performance of change detection. To improve the accuracy of change detection, a novel automatic SAR image change detection algorithm based on saliency detection and convolutional-wavelet neural networks is proposed. The log-ratio operator is adopted to generate the difference image, and the speckle reducing anisotropic diffusion is used to enhance the original multitemporal SAR images and the difference image. To reduce the influence of speckle noise, the salient area that probably belongs to the changed object is obtained from the difference image. The saliency analysis step can remove small noise regions by thresholding the saliency map, and interest regions can be preserved. Then an enhanced difference image is generated by combing the binarized saliency map and two input images. A hierarchical fuzzy c-means model is applied to the enhanced difference image to classify pixels into the changed, unchanged, and intermediate regions. The convolutional-wavelet neural networks are used to generate the final change map. Experimental results on five SAR data sets indicated the proposed approach provided good performance in change detection compared to state-of-the-art relative techniques, and the values of the metrics computed by the proposed method caused significant improvement.

An Advanced SAR Image Despeckling Method by Bernoulli-Sampling-Based Self-Supervised Deep Learning

As one of the main sources of remote sensing big data, synthetic aperture radar (SAR) can provide all-day and all-weather Earth image acquisition. However, speckle noise in SAR images brings a notable limitation for its big data applications, including image analysis and interpretation. Deep learning has been demonstrated as an advanced method and technology for SAR image despeckling. Most existing deep-learning-based methods adopt supervised learning and use synthetic speckled images to train the despeckling networks. This is because they need clean images as the references, and it is hard to obtain purely clean SAR images in real-world conditions. However, significant differences between synthetic speckled and real SAR images cause the domain gap problem. In other words, they cannot show superior performance for despeckling real SAR images as they do for synthetic speckled images. Inspired by recent studies on self-supervised denoising, we propose an advanced SAR image despeckling method by virtue of Bernoulli-sampling-based self-supervised deep learning, called SSD-SAR-BS. By only using real speckled SAR images, Bernoulli-sampled speckled image pairs (input–target) were obtained as the training data. Then, a multiscale despeckling network was trained on these image pairs. In addition, a dropout-based ensemble was introduced to boost the network performance. Extensive experimental results demonstrated that our proposed method outperforms the state-of-the-art for speckle noise suppression on both synthetic speckled and real SAR datasets (i.e., Sentinel-1 and TerraSAR-X).

Feature-transferable Pyramid Network for Cross-scale Object Detection in SAR Images

Multiscale object detection in Synthetic Aperture Radar (SAR) images can locate and recognize key objects in large-scene SAR images, and it is one of the key technologies in SAR image interpretation. However, for the simultaneous detection of SAR objects with large size differences, that is, cross-scale object detection, existing object detection methods are difficult to extract the features of cross-scale objects, and also difficult to realize cross-scale object simultaneous detection. In this study, we propose a multiscale object detection method based on the Feature-Transferable Pyramid Network (FTPN) for SAR images. In the feature extraction stage, the feature migration method is used to obtain an effective mosaic of the feature images of each layer and extract feature images with different scales. Simultaneously, the void convolution method is used to increase the receptive field of feature extraction and aid the network in extracting large object features. These steps can effectively preserve the features of objects of different sizes, to realize the simultaneous detection of cross-scale objects in SAR images. The experiments based on the GaoFen-3 SAR dataset, SAR Ship Detection Dataset (SSDD), and high-resolution SSDD-2.0 show that the proposed method can detect cross-scale objects, such as airports and ships in SAR images, and the mean Average Precision (mAP) can reach 96.5% on the existing dataset, which is 8.1% higher than that of the characteristic pyramid network algorithm. Moreover, the overall performance of the proposed method is better than that of the latest YOLOv4 and other object detection algorithms.

MMFF: Multi-manifold feature fusion based neural networks for target recognition in complex-valued SAR imagery

In this paper, we propose a novel multi-manifold feature fusion (MMFF) based deep learning framework for automatic target recognition (ATR) in complex-valued synthetic aperture radar (SAR) images. The conventional real-valued convolutional neural networks (CNN) usually ignore the intrinsic geometry structure of complex numbers, and the existing complex- and manifold-valued CNNs are not in accordance with the multimodality of SAR images. Founded on parallel manifold convolution branches for magnitude and phase data, the proposed MMFF model takes both the underlying geometry and the multimodal information of complex-valued SAR data into consideration, to learn manifold features of military and civilian targets for recognition tasks. Besides, we investigate manifold-valued layers such as activation function and batch normalization to leverage the MMFF architecture as well as the multimodal data. We validate our methods on two public SAR datasets in single- and dual-polarization, and achieve obviously better classification performance with less model parameters, compared with previously proposed real-, complex- and manifold-valued baselines.

Integration of the k-nearest neighbours and patch-based features for PolSAR image classification by using a two-branch residual network

ABSTRACT A two-branch residual deep learning network is proposed for polarimetric synthetic aperture radar (PolSAR) image classification in this work. The proposed method integrates local information of pixels from a rectangular region (patch-based features) with global information of the k-nearest neighbours (k-NN) constituting a deformable shape. Both contextual and polarimetric information are utilized to find the nearest neighbours. The experiments on two real Airborne Synthetic Aperture Radar (AIRSAR) datasets show superior performance of the proposed network with a relatively small training set.

Joint Inversion of Rupture across a Fault Stepover during the 8 August 2017 Mw 6.5 Jiuzhaigou, China, Earthquake

Abstract The 8 August 2017 Mw 6.5 Jiuzhaigou earthquake occurred in a tectonically fractured region in southwest China. We investigate the multifault coseismic rupture process by jointly analyzing teleseismic, strong-motion, high-rate Global Positioning System, and Interferometric Synthetic Aperture Radar (InSAR) datasets. We clearly identify two right-stepping fault segments and a compressional stepover based on variations in focal mechanisms constrained by coseismic InSAR deformation data. The average geometric parameters of the northwest and southeast segments are strike = 130°/dip = 57° and strike = 151°/dip = 70°, respectively. The rupture model estimated from a joint inversion of the seismic and geodetic datasets indicates that the rupture initiated on the southeastern segment and jumped to the northwestern segment, resulting in distinctive slip patches on the two segments. A 4-km-long coseismic slip gap was identified around the stepover, consistent with the aftershock locations and mechanisms. The right-stepping segmentation and coseismic rupture across the compressional stepover exhibited by the 2017 Jiuzhaigou earthquake are reminiscent of the multifault rupture pattern during the 1976 Songpan earthquake sequence farther south along the Huya fault system in three successive Ms∼7 events. Although the common features of fault geometry and stepover may control the similarity in event locations and focal mechanisms of the 2017 and 1976 sequences, the significantly wider (~15 km) stepover in the 1976 sequence likely prohibited coseismic rupture jumping and hence reduced seismic hazard.

A novel crop classification method based on ppfSVM classifier with time-series alignment kernel from dual-polarization SAR datasets

Rapid and accurate crop type mapping is of great significance for agricultural management and sustainable development. Time-series multi-polarization synthetic aperture radar (SAR) data is suitable for obtaining the large-scale distribution of crop types and continuously monitoring crops. At present, the classification method based on the time-series alignment of time-varying feature curves has been widely used, which can take the uncertainty of the phenological cycles of crops into account. The most classical method is the nearest neighbor (NN) classifier based on the dynamic time warping (DTW) alignment. While the DTW alignment does not consider the local shape of the curves and temporal ranges, and the NN classifier is inadequate in generalization, which restricts the accuracy of crop type mapping. In this paper, a pairwise proximity function support vector machine (ppfSVM) classification method with the time-weighted shapeDTW (TWshapeDTW) alignment is proposed. Firstly, the novel alignment method simultaneously considers the local shape of the curve and the temporal range of crops. Besides, the novel ppfSVM classifier with the time-series alignment kernel is established. Such kernel matrix considers dual-similarity metrics of multiple features, and it is positive semi-definite (PSD) in this classifier. With 42 Sentinel-1 dual-polarization SAR images located in Gansu Province, China, the crop classification maps in 2018 and 2019 are generated respectively. The proposed method in this paper obtains the overall accuracies of more than 90% in both two years, and the changes of crop types from 2018 to 2019 are also in line with the actual crop rotation. Compared with traditional classification methods (SVM and the NN method with the DTW alignment), it is found that the proposed method has higher overall accuracy (OA) and better robustness in the case of small number of samples. The OA's improvements of our method compared with the SVM method are 3% and 1% in 2018 and 2019, respectively. Such improvements are 14% and 12% respect to the NN method with the DTW alignment. This method can achieve more obvious improvement under the condition of less training samples and unaligned phenological sequences. Furthermore, the novel TWshapeDTW alignment is superior to the DTW and the time-weighted DTW alignment under the ppfSVM classifier. The OA's improvement introduced by the TWshapeDTW alignment respect to the DTW alignment can obtain 5% and 4% in 2018 and 2019, respectively.

A COMPARISON OF RANDOM FOREST AND LIGHT GRADIENT BOOSTING MACHINE FOR FOREST ABOVE-GROUND BIOMASS ESTIMATION USING A COMBINATION OF LANDSAT, ALOS PALSAR, AND AIRBORNE LIDAR DATA

Abstract. Sustainable forest management is a critical topic which contributes to ecological, economical, and socio-cultural aspect of the environment. Providing accurate AGB maps is of paramount importance for sustainable forest management, carbon accounting, and climate change monitoring. The main goal of this study was to leverage the potential of two machine learning algorithms for predicting AGB using optical and synthetic aperture radar (SAR) datasets. To achieve this goal random forest (RF) and light gradient boosting machine (LightGBM) models were deployed to predict AGB values in Huntington Wild Forest (HWF) in Essex County, NY using continuous forest inventory (CFI) plots. Both models were trained and evaluated based on airborne light detection and ranging (LiDAR) data, Landsat imagery, advanced land observing satellite (ALOS) phased array type L-band Synthetic Aperture Radar (PALSAR), and their combination. The integration of airborne LiDAR, optic, and SAR datasets provided the best results in terms of root mean square error (RMSE) and mean bias error (MBE). The RF model outperformed the LightGBM in all scenarios (LiDAR, Landsat 5, ALOS PALSAR, and their combination). The RF model was able to predict AGB values with the RMSE of 51.90 Mg/ha and MBE of −0.189 Mg/ha for the combination of LiDAR, optic, and SAR data, while LightGBM estimated the AGB values with the RMSE of 52.78 Mg/ha and MBE of −0.253 Mg/ha. LightGBM is more sensitive to noise and there are lots of hyperparameters that need to be tuned which highly affect its performance.

Evaluation of the change in synthetic aperture radar imaging using transfer learning and residual network

Change detection from synthetic aperture radar images becomes a key technique to detect change area related to some phenomenon as flood and deformation of the earth surface. This paper proposes a transfer learning and Residual Network with 18 layers (ResNet-18) architecture-based method for change detection from two synthetic aperture radar images. Before the application of the proposed technique, batch denoising using convolutional neural network is applied to the two input synthetic aperture radar image for speckle noise reduction. To validate the performance of the proposed method, three known synthetic aperture radar datasets (Ottawa; Mexican and for Taiwan Shimen datasets) are exploited in this paper. The use of these datasets is important because the ground truth is known, and this can be considered as the use of numerical simulation. The detected change image obtained by the proposed method is compared using two image metrics. The first metric is image quality index that measures the similarity ratio between the obtained image and the image of the ground truth, the second metrics is edge preservation index, it measures the performance of the method to preserve edges. Finally, the method is applied to determine the changed area using two Sentinel 1 B synthetic aperture radar images of Eddahbi dam situated in Morocco.

Target Detection Network for SAR Images Based on Semi-Supervised Learning and Attention Mechanism

The existing Synthetic Aperture Radar (SAR) image target detection methods based on convolutional neural networks (CNNs) have achieved remarkable performance, but these methods require a large number of target-level labeled training samples to train the network. Moreover, some clutter is very similar to targets in SAR images with complex scenes, making the target detection task very difficult. Therefore, a SAR target detection network based on a semi-supervised learning and attention mechanism is proposed in this paper. Since the image-level label simply marks whether the image contains the target of interest or not, which is easier to be labeled than the target-level label, the proposed method uses a small number of target-level labeled training samples and a large number of image-level labeled training samples to train the network with a semi-supervised learning algorithm. The proposed network consists of a detection branch and a scene recognition branch with a feature extraction module and an attention module shared between these two branches. The feature extraction module can extract the deep features of the input SAR images, and the attention module can guide the network to focus on the target of interest while suppressing the clutter. During the semi-supervised learning process, the target-level labeled training samples will pass through the detection branch, while the image-level labeled training samples will pass through the scene recognition branch. During the test process, considering the help of global scene information in SAR images for detection, a novel coarse-to-fine detection procedure is proposed. After the coarse scene recognition determining whether the input SAR image contains the target of interest or not, the fine target detection is performed on the image that may contain the target. The experimental results based on the measured SAR dataset demonstrate that the proposed method can achieve better performance than the existing methods.

SAR image change detection based on sparse representation and a capsule network

ABSTRACT Synthetic aperture radar (SAR) image change detection has good application prospects in the domain of remote sensing image processing. However, the accuracy of change detection is negatively affected by the inherent speckle noise in SAR images. To solve this problem, a method based on sparse representation (SR) and a capsule network is proposed. Firstly, sparse features of the difference image (DI) are extracted by the SR method. Secondly, a lightweight capsule network (L-CapsNet) is constructed, which is used to mine the spatial relationship between features. The network classifies the changed and unchanged pixels. Finally, the change map (CM) is generated. The proposed method can obtain more robust features while reducing the influence of speckle noise. Experiments are performed on four SAR data sets to compare the proposed method with related methods. The results confirm the superior performance of the proposed method.