Synthetic Aperture Radar Datasets Research Articles

SLIC Superpixel Segmentation for Polarimetric SAR Images

Superpixel segmentation approaches for polarimetric synthetic aperture radar (SAR) images have only been studied in recent years. Simple linear iterative clustering (SLIC) is a simple and efficient superpixel segmentation method, first proposed for optical images. It basically includes three implementation steps, i.e., initialization, local <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, and postprocessing. The challenge of applying SLIC to polarimetric SAR images lies in constructing the effective spatial and feature similarity and proposing the efficient segmentation procedure. In this study, to address both issues, we modify the SLIC clustering function to adapt the characteristics of polarimetric statistical measures. A new initialization method is proposed, which exploits the image gradient information to produce robust cluster centers. Furthermore, in an effort to give a comprehensive comparison and provide a fair assessment of the feature similarities for polarimetric SAR imagery, four classic statistical distances, among which two were not studied along with the SLIC previously, are embedded in the modified clustering function. The proposed method is validated by comparing with state-of-the-art SLIC-based algorithms and also the Ncut and TurboPixel algorithms. Experiments on extensive polarimetric SAR data sets show that the proposed method can significantly improve the segmentation results, with producing better boundary adherence and compact as well as uniform superpixels. We also obtain distinct conclusions that are different from the existing studies when investigating the performances of the statistical measures.

Multi-Block Mixed Sample Semi-Supervised Learning for SAR Target Recognition

In recent years, synthetic aperture radar (SAR) automatic target recognition has played a crucial role in multiple fields and has received widespread attention. Compared with optical image recognition with massive annotation data, lacking sufficient labeled images limits the performance of the SAR automatic target recognition (ATR) method based on deep learning. It is expensive and time-consuming to annotate the targets for SAR images, while it is difficult for unsupervised SAR target recognition to meet the actual needs. In this situation, we propose a semi-supervised sample mixing method for SAR target recognition, named multi-block mixed (MBM), which can effectively utilize the unlabeled samples. During the data preprocessing stage, a multi-block mixed method is used to interpolate a small part of the training image to generate new samples. Then, the new samples are used to improve the recognition accuracy of the model. To verify the effectiveness of the proposed method, experiments are carried out on the moving and stationary target acquisition and recognition (MSTAR) data set. The experimental results fully demonstrate that the proposed MBM semi-supervised learning method can effectively address the problem of annotation insufficiency in SAR data sets and can learn valuable information from unlabeled samples, thereby improving the recognition performance.

A fast, edge-preserving, distance-regularized model with bilateral filtering for oil spill segmentation of SAR images

Marine oil spills are among the most significant sources of marine pollution. Synthetic aperture radar (SAR) has been used to improve oil spill observations because of its advantages in oil spill detection and identification. However, speckle noise, weak boundaries, and intensity inhomogeneity often exist in the oil spill regions of SAR imagery, which will seriously affect the accurate identification of oil spills. To enhance marine oil spill segmentation of SAR images, a fast, edge-preserving framework based on the distance-regularized level set evolution (DRLSE) model was proposed. Specifically, a bilateral filter penalty term is designed and incorporated into the DRLSE energy function (BF-DRLSE) to preserve the edges of oil spills, and an adaptive initial box boundary was selected for the DRLSE model to reduce the operation time complexity. Two sets of RadarSat-2 SAR data were used to test the proposed method. The experimental results indicate that the bilateral filtering scheme incorporated into the energy function during level set evolution improved the stability of level set evolution. Compared with other methods, the proposed improved BF-DRLSE algorithm displayed a higher overall segmentation accuracy (97.83%). In addition, using an appropriate initial box boundary for the DRLSE method accelerated the global search process, improved the accuracy of oil spill segmentation, and reduced computational time. Therefore, the results suggest that the proposed framework is effective and applicable for marine oil spill segmentation.

Robust unsupervised small area change detection from SAR imagery using deep learning

Small area change detection using synthetic aperture radar (SAR) imagery is a highly challenging task, due to speckle noise and imbalance between classes (changed and unchanged). In this paper, a robust unsupervised approach is proposed for small area change detection using deep learning techniques. First, a multi-scale superpixel reconstruction method is developed to generate a difference image (DI), which can suppress the speckle noise effectively and enhance edges by exploiting local, spatially homogeneous information. Second, a two-stage centre-constrained fuzzy c-means clustering algorithm is proposed to divide the pixels of the DI into changed, unchanged and intermediate classes with a parallel clustering strategy. Image patches belonging to the first two classes are then constructed as pseudo-label training samples, and image patches of the intermediate class are treated as testing samples. Finally, a convolutional wavelet neural network (CWNN) is designed and trained to classify testing samples into changed or unchanged classes, coupled with a deep convolutional generative adversarial network (DCGAN) to increase the number of changed class within the pseudo-label training samples. Numerical experiments on four real SAR datasets demonstrate the validity and robustness of the proposed approach, achieving up to 99.61% accuracy for small area change detection.

Analyzing the Separability of SAR Classification Dataset in Open Set Conditions

In synthetic aperture radar (SAR) image classification applications, there are three categories of data, including training and benchmark data with fixed classes, as well as actual data in practical applications. A real problem comes that there exist unknown classes not included in training and benchmark data, which is defined as the open set condition. However, little work on recognizing unknown classes and analyzing the separability of SAR datasets has been developed. Motivated by this observation, this article demonstrates the difficulty of practical classification and analyzes SAR dataset separability in open set conditions. In this article, the SAR separability analyzer (SAR-SA) is proposed to model each known class as a multivariate Gaussian distribution. SAR-SA can classify the known classes and recognize the samples locating in each known distribution with low probabilities as unknown. Besides, SAR datasetwise separability index (DSI) and classwise separability index (CSI) are defined to quantify the separability in open set conditions at the dataset level and class level. DSI and CSI are effective indicators of the difficulty of SAR classification datasets. Extensive experimental results demonstrate that the DSI in open set conditions is nearly half of that in supervised conditions. Dataset with low DSI is hard to realize accurate classification in open set conditions. At the class level, even though the SAR image classes are semantically different from each other, there exists more or less overlap between the distributions of supervised known classes and unknown classes. Classes with low CSI are harder to be correctly classified and recognized.

Multitask Learning for Ship Detection From Synthetic Aperture Radar Images

Ship detection from synthetic aperture radar (SAR) images is inherently subject to the special imaging mechanism of SAR. In recent years, deep-learning-based techniques for detecting objects from optical images have rapidly advanced and promoted the development of SAR image detection technology. However, the strong speckle noise in SAR images degrades low-level feature learning in shallow layers, hindering the higher level learning of semantic features for object detection. In view of the problems encountered in direct end-to-end feature learning for object detection and the close relationship between objects and auxiliary cues, a multitask learning-based object detector (MTL-Det) is proposed in this article to distinguish ships in SAR images. The proposed approach models the ship detection problem, not as a single object detection task, but as three cooperative tasks. The model involves two auxiliary subtasks that are focused on learning object-specific cues ( e.g ., texture and shape) for the ship detection task, which is constrained by the pseudoground truth generated by the main task. Assisted by auxiliary subtasks, the low-level features are robust to speckle noise and reliably support high-level feature learning. Compared with traditional single-task-based object detectors, more discriminative object-specific features are learned by multitask learning without the extra cost of manual labeling. The experiments conducted in this study help demonstrate the advantages of MTL-Det in improving the ship detection performance on two SAR datasets: high-resolution SAR images dataset and large-scale SAR ship detection dataset-v1.0.

A Benchmark High-Resolution GaoFen-3 SAR Dataset for Building Semantic Segmentation

Deep learning is increasingly popular in remote sensing communities and already successful in land cover classification and semantic segmentation. However, most studies are limited to the utilization of optical datasets. Despite few attempts applied to synthetic aperture radar (SAR) using deep learning, the huge potential, especially for the very high resolution (VHR) SAR, are still underexploited. Taking building segmentation as an example, the VHR SAR datasets are still missing to the best of our knowledge. A comparable baseline for SAR building segmentation does not exist, and which segmentation method is more suitable for SAR image is poorly understood. This article first provides a benchmark high-resolution (1 m) GaoFen-3 SAR datasets, which cover nine cities from seven countries, review the state-of-the-art semantic segmentation methods applied to SAR, and then summarize the potential operations to improve the performance. With these comprehensive assessments, we hope to provide the recommendation and roadmap for future SAR semantic segmentation.

Multi-Scale Ship Detection From SAR and Optical Imagery Via A More Accurate YOLOv3

Deep learning detection methods use in ship detection remains a challenge, owing to the small scale of the objects and interference from complex sea surfaces. In addition, existing ship detection methods rarely verify the robustness of their algorithms on multisensor images. Thus, we propose a new improvement on the “you only look once” version 3 (YOLOv3) framework for ship detection in marine surveillance, based on synthetic aperture radar (SAR) and optical imagery. First, improved choices are obtained for the anchor boxes by using linear scaling based on the k-means++ algorithm. This addresses the difficulty in reflecting the advantages of YOLOv3's multiscale detection, as the anchor boxes of a single detection target type between different detection scales have small differences. Second, we add uncertainty estimators for the positioning of the bounding boxes by introducing a Gaussian parameter for ship detection into the YOLOv3 framework. Finally, four anchor boxes are allocated to each detection scale in the Gaussian-YOLO layer instead of three as in the default YOLOv3 settings, as there are wide disparities in an object's size and direction in remote sensing images with different resolutions. Applying the proposed strategy to "YOLOv3-spp” and "YOLOv3-tiny,” the results are enhanced by 2%-3%. Compared with other models, the improved-YOLOv3 has the highest average precision on both the optical (93.56%) and SAR (95.52%) datasets. The improved-YOLOv3 is robust, even in the context of a mixed dataset of SAR and optical images comprising images from different satellites and with different scales.

AN APPROACH TO POSITIONAL QUALITY CONTROL METHODS FOR AIRBORNE INSAR HIGH-RESOLUTION X-BAND ORTHOIMAGES AND P-BAND DIGITAL TERRAIN MODEL

Abstract: The positional validation of datasets is an important step for cartography studies since it allows learning about its accuracy, and also indicates the data process quality. However, the positional validation of Synthetic Aperture Radar (SAR) datasets have some additional challenges when compared to optical images due to the geometric distortions. We employ existing targets such as traffic signs and lampposts in the scene and identify them on the image as control points. We performed the validation of the geographic coordinates used as planialtimetric positional control points, using both the amplitude backscattering orthoimage and the Digital Terrain Model (DTM) generated from the InSAR system. We employed the NMAS, ASPRS and NSSDA tests along with information by the Brazilian Standards. This validation showed these control points presented the following results for 1:10,000 scale: NMAS test - class “A” in PEC and PEC-PCD; ASPRS test - RMSE x = 1.317m, RMSE y = 1.231m and RMSE z = 1.145m; and NSSDA test - RMSE r = 1,802m, Precision r = 3.118m and Precision z = 2.244m. These results prove we can use the proposed targets as control points and the used InSAR datasets meet the expected quality for generation of geotechnic products for 1:10,000 scale.

LCS-EnsemNet: A Semisupervised Deep Neural Network for SAR Image Change Detection With Dual Feature Extraction and Label-Consistent Self-Ensemble

Change detection (CD) in synthetic aperture radar (SAR) images faces two challenging problems: 1) inherent speckle noise in SAR data causes the class confusion and affects the image understanding for inferring category of each image pixel; and 2) adequate labeled samples are quite hard to collect, which is the major limitation for supervised methods. In this article, we develop a novel deep learning-based semisupervised method to address these challenges. The method firstly incorporates a pixel-wise log-ratio difference image (DI) and its saliency map to produce a spatially enhanced (SE) DI using a reweighting scheme based on the fact that changed pixels exhibit higher saliency than unchanged pixels. As a result, prominent changed regions are highlighted and the interclass separability is increased. We construct pixel-wise and context-wise features based on the log-ratio DI and SE DI to jointly express the dissimilarity at each pixel. Second, a two-branch deep neural network taking the dual features as input is devised to reveal and refine underlying label information from abundant unlabeled samples with the help of a two-stage refinement strategy for improving generalization ability, called LCS-EnsemNet. The refinement strategy, namely label-consistent self-ensemble, exploits label consistency between dual features and across multiple classifiers to ensure the reliability of pseudo labels. Finally, the cross-entropy loss is calculated with the limited labeled data and selected pseudo labeled samples to optimize the LCS-EnsemNet. The proposed method is evaluated on three low/medium- resolution and one high-resolution SAR data sets, and experimental results have demonstrated its effectiveness.

Detection of Oil Spill Using SAR Imagery Based on AlexNet Model.

Synthetic aperture radar (SAR) plays an irreplaceable role in the monitoring of marine oil spills. However, due to the limitation of its imaging characteristics, it is difficult to use traditional image processing methods to effectively extract oil spill information from SAR images with coherent speckle noise. In this paper, the convolutional neural network AlexNet model is used to extract the oil spill information from SAR images by taking advantage of its features of local connection, weight sharing, and learning for image representation. The existing remote sensing images of the oil spills in recent years in China are used to build a dataset. These images are enhanced by translation and flip of the dataset, and so on and then sent to the established deep convolutional neural network for training. The prediction model is obtained through optimization methods such as Adam. During the prediction, the predicted image is cut into several blocks, and the error information is removed by corrosion expansion and Gaussian filtering after the image is spliced again. Experiments based on actual oil spill SAR datasets demonstrate the effectiveness of the modified AlexNet model compared with other approaches.

Speckle-Free SAR Image Ship Detection.

Ship detection is one of important applications for synthetic aperture radar (SAR). Speckle effects usually make SAR image understanding difficult and speckle reduction becomes a necessary pre-processing step for majority SAR applications. This work examines different speckle reduction methods on SAR ship detection performances. It is found out that the influences of different speckle filters are significant which can be positive or negative. However, how to select a suitable combination of speckle filters and ship detectors is lack of theoretical basis and is also data-orientated. To overcome this limitation, a speckle-free SAR ship detection approach is proposed. A similar pixel number (SPN) indicator which can effectively identify salient target is derived, during the similar pixel selection procedure with the context covariance matrix (CCM) similarity test. The underlying principle lies in that ship and sea clutter candidates show different properties of homogeneity within a moving window and the SPN indicator can clearly reflect their differences. The sensitivity and efficiency of the SPN indicator is examined and demonstrated. Then, a speckle-free SAR ship detection approach is established based on the SPN indicator. The detection flowchart is also given. Experimental and comparison studies are carried out with three kinds of spaceborne SAR datasets in terms of different polarizations. The proposed method achieves the best SAR ship detection performances with the highest figures of merits (FoM) of 97.14%, 90.32% and 93.75% for the used Radarsat-2, GaoFen-3 and Sentinel-1 datasets, accordingly.

Synthetic Aperture Radar Image Change Detection via Siamese Adaptive Fusion Network

Synthetic aperture radar (SAR) image change detection is a critical yet challenging task in the field of remote sensing image analysis. The task is nontrivial due to the following challenges: First, intrinsic speckle noise of SAR images inevitably degrades the neural network because of error gradient accumulation. Furthermore, the correlation among various levels or scales of feature maps is difficult to be achieved through summation or concatenation. Toward this end, we proposed a siamese adaptive fusion (AF) network for SAR image change detection. To be more specific, two-branch CNN is utilized to extract high-level semantic features of multitemporal SAR images. Besides, an AF module is designed to adaptively combine multiscale responses in convolutional layers. Therefore, the complementary information is exploited, and feature learning in change detection is further improved. Moreover, a correlation layer is designed to further explore the correlation between multitemporal images. Thereafter, robust feature representation is utilized for classification through a fully connected layer with softmax. Experimental results on four real SAR datasets demonstrate that the proposed method exhibits superior performance against several state-of-the-art methods. Our codes are available at https://github.com/summitgao/SAR_CD_SAFNet .

Shape Similarity Intersection-Over-Union Loss Hybrid Model for Detection of Synthetic Aperture Radar Small Ship Objects in Complex Scenes

With the continuous development and utilization of marine environments, the demand for accurate identification of ship targets at sea is increasing in both military and civilian fields. Synthetic aperture radar (SAR) is used to detect ship targets at sea and can provide 24-h detection under any weather conditions. Deep-learning models enable the effective detection of ship targets using SAR images; however, the recognition accuracy may be low or false positives may occur in complex scenarios wherein it is difficult to detect the ship targets. Current target-detection tasks include target classification and positioning through bounding-box regression. Herein, a regression loss function is derived to calculate the position of the bounding box, and intersection over union (IoU) is applied to estimate the positioning accuracy. As a result, a gap exists between the commonly used positioning losses for regressing the parameters of a bounding box and the optimization of these metric values. Therefore, the proposed hybrid model combines classification, localization, and segmentation with a novel multitask loss function for boundary-box localization based on the improved IoU. This solves the problem of inconsistency between training and evaluation and improves the positioning accuracy. Experiments were conducted using the SAR dataset for ship detection; the dataset was labeled by SAR experts and included multiscale ship chips with a resolution of 256 pixels in both range and azimuth. In summary, the experimental results indicate that the proposed hybrid model could improve the detection accuracy in complex scenarios, and its false-positive rate is significantly lower than those of the other models.

Model-Based Interpretation of PolSAR Data for the Characterization of Glacier Zones in Greenland

Different approaches have been proposed in recent years to map glacier zones using Synthetic Aperture Radar (SAR). Most of them rely on empirical or statistical approaches for the interpretation of backscatter intensity measurements and their seasonal variations related to the melting regime of the different glacier zones. This study focuses on the potential of polarimetric SAR measurements to characterize the scattering in the different glacier zones in order to distinguish between them. For this, a general two component scattering model is proposed, accounting for surface and sub-surface contributions in the different glacier zones. The (anisotropic) propagation through snow, firn and ice layers is also accounted for as it plays an important role in the interpretation of the polarimetric signatures. A set of five polarimetric descriptors is proposed to describe the projection of the model on the SAR measurements. The ability of the model to characterize the scattering in the different glacier zones is validated and discussed at different frequencies (X-, C-, and L-band). For this, a multi-frequency airborne SAR dataset acquired by the DLRs F-SAR sensor in West Greenland during the ARCTIC 2015 campaign is used. The achieved results indicate that L-band polarimetric measurements allow to differentiate the different glacier zones based on the sub-surface (e.g. volume) scattering characteristics. In contrast, the increased (back-) scattering complexity and the reduced penetration capability limit the potential of X- and C-band polarimetric measurements.

Change Detection in Hyperdimensional Images Using Untrained Models

Deep transfer-learning based change detection methods are dependent on the availability of sensor-specific pre-trained feature extractors. Such feature extractors are not always available due to lack of training data, especially for hyperspectral sensors and other hyperdimensional images. Moreover models trained on easily available multispectral (RGB/RGB-NIR) images cannot be reused on such hyperdimensional images due to their irregular number of bands. While hyperdimensional images show large number of spectral bands, they generally show much less spatial complexity, thus reducing the requirement of large receptive fields of convolution filters. Recent works in the computer vision have shown that even untrained deep models can yield remarkable result in some tasks like super-resolution and surface reconstruction. This motivates us to make a bold proposition that untrained lightweight deep model, initialized with some weight initialization strategy, can be used to extract useful semantic features from bi-temporal hyperdimensional images. Based on this proposition, we design a novel change detection framework for hyperdimensional images by extracting bi-temporal features using an untrained model and further comparing the extracted features using Deep Change Vector Analysis to distinguish changed pixels from the unchanged ones. We further use the deep change hypervectors to cluster the changed pixels into different semantic groups. We conduct experiments on four change detection datasets: three hyperspectral datasets and a hyperdimensional Polarimetric Synthetic Aperture Radar dataset. The results clearly demonstrate that the proposed method is suitable for change detection in hyperdimensional remote sensing data.

Performance evaluation of multi-parametric synthetic aperture radar data for geological lineament extraction

ABSTRACT The demarcation of linear geomorphological features due to fracture systems, underlying or exposed faults, drainage lines etc. from satellite images has been one of the most widely used applications in geological studies. Synthetic Aperture Radar (SAR) data has been recognized as an effective tool for lineament extraction due to its unique sensitivity to subtle topography, soil moisture and texture. However, the typical SAR sensor parameters such as frequency, polarization and look direction plays a vital role in enhancing the performance for a specific application. Therefore, the study aims to analyse the multi-frequency (C- and L-band), multi-polarization (like and cross) and multi-look direction SAR data for geological lineaments extraction. In this paper, Sentinel-1 and Advanced Land Observation Satellite (ALOS)-2 Phased Array type L-band Synthetic Aperture Radar (PALSAR)-2 datasets acquired over a part of the Indo-Burma ranges of Manipur region (northeastern part of India) are considered for analysis. The lineaments extracted from SAR datasets are compared with reference to lineaments provided by the Geological Survey of India (GSI) through their portal. The study observes that the like-polarization C-band SAR data provide optimal results for this application. It is also observed that the integration of lineaments extracted from multi-look SAR data can significantly improve the performance than that derived from single-look data. Overall, the study also focuses on the spatial distribution of lineaments with respect to the local structural regime for a better understanding of the regional tectonic setup.

New empirical backscattering models for estimating bare soil surface parameters

ABSTRACT Various models have been proposed to estimate the degree of backscatter in Synthetic Aperture Radar (SAR) images. However, it is still necessary to calibrate these models based on the characteristics of different study areas and to propose new models to achieve the highest possible accuracy in estimating the backscattering coefficient () SAR. In this study, three empirical models, including Champion, Sahebi and Zribi/Dechambre, were initially calibrated for two SAR datasets (i.e. The Airborne Synthetic Aperture Radar (AIRSAR) and Canadian Space Agency radar satellite (RADARSAT-1)) acquired over two bare soil study areas with various soil characteristics. The Zribi/Dechambre model was then modified by revising the roughness parameter to obtain higher accuracy in estimating over a larger range of incidence angles (θ). A new empirical model was also proposed by combining the four parameters of Soil Moisture (SM), standard deviation of surface height -root mean square- (rms), correlation length (l), and θ. To this end, the most appropriate form of the regression model was investigated and used for each of these parameters to obtain the highest correlation between the in-situ data and values. A comparison of the empirical models showed that the modified Zribi/Dechambre had the highest accuracy in predicting values with the Root Mean Square Errors (RMSE) of 1.20 dB and 1.59 dB over Oklahoma and Quebec, respectively. Furthermore, coefficients values of the new proposed model remained stable in the two datasets unlike the other investigated models. In this study, the effects of l on the accuracy of the new proposed model were also assessed. It was concluded that l had a considerable impact on the accuracy of the proposed model and including this parameter can improve the accuracy by up to 1 dB.

SAR and LIDAR Datasets for Building Damage Evaluation Based on Support Vector Machine and Random Forest Algorithms—A Case Study of Kumamoto Earthquake, Japan

The evaluation of buildings damage following disasters from natural hazards is a crucial step in determining the extent of the damage and measuring renovation needs. In this study, a combination of the synthetic aperture radar (SAR) and light detection and ranging (LIDAR) data before and after the earthquake were used to assess the damage to buildings caused by the Kumamoto earthquake. For damage assessment, three variables including elevation difference (ELD) and texture difference (TD) in pre- and post-event LIDAR images and coherence difference (CD) in SAR images before and after the event were considered and their results were extracted. Machine learning algorithms including random forest (RDF) and the support vector machine (SVM) were used to classify and predict the rate of damage. The results showed that ELD parameter played a key role in identifying the damaged buildings. The SVM algorithm using the ELD parameter and considering three damage rates, including D0 and D1 (Negligible to slight damages), D2, D3 and D4 (Moderate to Heavy damages) and D5 and D6 (Collapsed buildings) provided an overall accuracy of about 87.1%. In addition, for four damage rates, the overall accuracy was about 78.1%.

Polarimetric SAR Decomposition by Incorporating a Rotated Dihedral Scattering Model

In this letter, we propose a new scattering model to describe the polarimetric scattering information of the real part of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$T_{23}$ </tex-math></inline-formula> in the coherency matrix. To achieve this goal, by combining the dihedral corner reflector scattering model and the polarimetric orientation angle (POA), a rotated dihedral scattering model is proposed. The proposed model is embedded into Singh’s six-component decomposition model, and we further develop a seven-component decomposition model. The proposed method was validated by polarimetric synthetic aperture radar (SAR) data sets acquired by the ALOS-2/PALSAR-2 and AIRSAR systems. The results show that, compared with the existing decomposition methods, the proposed method has a superior ability to distinguish oriented buildings from vegetation.