Training of synthetic aperture radar (SAR) target detection and recognition methods based on deep learning heavily relies on a large amount of data. As one of the significant approaches to address the scarcity of SAR data, SAR image intelligent generation methods have witnessed rapid development. However, these methods often require many data samples for learning and are prone to deviating from the physical scattering characteristics. To address these issues, this paper proposes a two-stage SAR image generation method based on attribute feature decoupling within a generative adversarial network (GAN) architecture. In the first stage, the original SAR target image undergoes feature extraction and reconstruction, yielding generated images highly similar to real images. The attribute features decoupled during this process correlate with the scattering characteristics of SAR target, providing guiding information for generating target images in the second stage. In the second stage, by applying perturbations to specific dimensions of the decoupled features, we can reconstruct target images with altered attributes, achieving diverse data augmentation. Multi-task discrimination based on pixel intensity, authenticity, and feature distance differences enhances the quality of generated images across multiple levels. The decoupled representation-driven generation paradigm simplifies the network’s mapping learning task through task decomposition, diminishing the dependency on the volume of data. The experimental results demonstrate that the generated images possess higher quality and superior application performance, with an improvement of 5.23% in recognition accuracy.

Large-scale, diverse and high-quality data is the foundation and key to achieving good generalization in target detection and recognition for deep learning-based algorithms. Directly collecting synthetic aperture radar (SAR) image data faces the difficulty in acquisition and high costs. Traditional SAR image simulation methods are limited by geometric and electromagnetic computation errors in their modeling process, and the high computational burden as well. Generative adversarial networks (GANs) offer a new approach for SAR image generation, but they struggle to achieve satisfactory results in terms of image quality and diversity. In order to overcome this problem, we propose a new type of GAN to learn the spatial relationship of the targets more effectively. Taking the real SAR images as input, we extract the target information through the capsule network, perturb the extracted features and adopt the attention mechanism to improve the quality and diversity of the augmented data.

Although remote sensing image registration has been developing at a high speed for decades, polarimetric synthetic aperture radar (PolSAR) image registration is still a challenging task because of the presence of polarimetric scattering differences, geometric distortions, and speckle noise. Due to the lack of PolSAR image training data, the generalization performance of deep learning-based registration methods is poor and cannot fundamentally solve the problem of PolSAR image registration. In this article, we propose a novel PolSAR image registration framework that integrates feature selection, feature descriptor extraction, and template matching. First, we use structural similarity (SSIM) to select polarimetric features that are similar in structural information, thus using structural information to overcome polarimetric scattering differences. On this basis, a feature descriptor named oriented gradient of polarimetric feature (OGPF) is proposed to overcome the polarimetric scattering information difference by extracting geometric structure information using oriented gradient channels (OGCs) and 3-D Gaussian convolution. Finally, we propose to use polarimetric whitening filter (PWF) and nonmaximum suppression (NMS) to extract keypoints with significant structural information to reduce the interference of speckle noise on keypoint selection, and further propose a template downsampling strategy to reduce the complexity of template matching. The proposed method is evaluated using six pairs of PolSAR images with different scenes, and the results show that its registration performance outperforms the state-of-the-art methods.

Polarimetric synthetic aperture radar (PolSAR) is an active microwave imaging system. Due to the coherence characteristic of PolSAR imaging, inherent coherent speckle noise exists in PolSAR images. The registration of PolSAR images is severely affected by speckle noise. Therefore, we first propose a joint filter that combines Refined-Lee filtering and polarimetric whitening filtering (PWF). The filter first applies Refined-Lee filtering to PolSAR images, which greatly reduces the speckle noise while maintaining high-resolution detailed information of the texture, and then uses PWF to normalize and whiten the polarimetric matrix to further limit the interference of speckle noise. After that, the binary robust invariant scalable keypoints (BRISK) algorithm is used to extract high-quality keypoints from the denoised PolSAR image. Then a novel Siamese Multiscale Attention Network (SMAN) is designed, which uses attention modules to construct feature descriptors with different scales. To fully utilize polarimetric information, we adopt the polarimetric covariance matrix and three polarimetric features as inputs to the network and the Second Order Similarity (SOS) as the loss function to train the network. In the keypoint matching stage, we present to use the symmetric displacement distance to further constrain the keypoint pairs obtained by the initial matching, which improves the accuracy of matching keypoint pairs. Experimental results show that our proposed method can effectively reduce the interference of speckle noise and overcome non-linear differences, geometric distortions, and differences in polarimetric scattering information to achieve accurate PolSAR image registration.

When the geometric distortion of the SAR images to be registered is large, the spatial correspondence between the feature points of the two images will change significantly. Hence, the registration of SAR images with large geometric distortion is challenging. To solve this problem, a two-stage registration method of SAR images with large distortion based on superpixel segmentation is proposed in this paper. Firstly, the two SAR images are coarsely registered by geographic coordinate referencing. After coarse registration, superpixel segmentation is performed on the two SAR images respectively. Next, in the superpixel neighborhood of the reference image, we slide the corresponding superpixel template of the sensed image, finding its position with the highest similarity in the reference image. Compared with the traditional fixed-size template, the superpixel template can segment the distorted region more effectively. Meanwhile, with the help of the adaptive threshold detector proposed in this paper, the regions with varying degrees of distortion can be distinguished based on the similarity. Further, the geometry mapping relationship is calculated for the regions with different distortion degrees in the images respectively, and the corresponding feature points in different images are accurately matched to complete the fine registration. Finally, the registration results of regions with different degrees of distortion are fused to obtain the final SAR image registration results. Experimental results based on Sentinel-1 data show that the registration accuracy of the proposed method can reach within 1 pixel.

Ship detection in synthetic aperture radar (SAR) images is critical to ocean surveillance and rescue. Although many deep learning SAR ship detection methods have been proposed, the performance of these methods depends on the size and quality of the training samples. To resolve these issues, this letter presents an unsupervised ship detection method in SAR images using superpixel segmentation and cross stage partial network (CSPNet). First, the SAR image is over-segmented into superpixels based on our previously proposed superpixel generation algorithm. Then, the complex signal kurtosis (CSK) and a local superpixel contrast are integrated as a statistical indicator for the automatic identification of ship superpixels and background superpixels, thus leading to generation of training samples. Finally, the segmented superpixels are input to the CSPNet, which can learn a representative feature set with high discrimination ability between ships and backgrounds. Our method can achieve pixel-level detection map rather than the bounding box result. Experiments based on the Gaofen-3 and TerraSAR SAR data demonstrate that our method can achieve above 90% actual detection rate.

Small target detection is a very challenging problem since a small target contains only a few pixels in size. At present, many deep learning-based detection algorithms for small targets have achieved remarkable results, mainly including improvements in data augmentation, multiscale images, multiscale features, training strategies, and so on. However, these deep learning-based methods cannot select the positive and negative samples for the large-difference-scale targets well in the label assignment operation. The reason is that intersection over union (IoU), which is widely used in the most target detection networks, has great limitations for small target detection. However, in practical applications, there are often some large-difference-scale targets in synthetic aperture radar (SAR) images, especially existing some tiny targets due to the limitation of resolution. To fundamentally break the limitations of IoU, we propose to use the Bhattacharyya distance (BD) instead of the IoU metric to improve the performance of small target detection. We further revise the Bhattacharyya distance (RBD) to better measure the deviation of bounding boxes for targets with large differences in size. RBD can embed anchor-based detectors to replace the IoU metric in label assignment and nonmaximum suppression (NMS). The proposed method is evaluated on the LS-SSDD-v1.0 dataset and the experimental results show that the proposed method outperforms the state-of-the-art methods.

Polarimetric features extracted from the polarimetric synthetic aperture radar data contain a wealth of target scattering information, but usually lead to the problems, such as network learning burden and high computational consumption. A multichannel fusion convolutional neural network based on scattering mechanisms was presented in this letter. First, the polarimetric features were divided into three categories according to their corresponding scattering mechanisms, and put into three network channels, respectively. Second, a new feature output was constructed based on the fusion of three-channel output features. Third, the four output features were cascaded through two fully connected layers and the Softmax classifier to get the classification result. Moreover, a new loss function was defined, combining cross entropy and average cross entropy to prevent network overfitting. Experimental results on airborne synthetic aperture radar (AIRSAR) and GF-3 data set verified the effectiveness of the proposed method in the aspect of classification accuracy and small sample.

Time-series polarimetric synthetic aperture radar (PolSAR) has been proven to be an effective technique for crop classification and agricultural activity monitoring. However, the characterization and utilization of time-series PolSAR data by existing methods are still inadequate. They are unable to extract and utilize time-varying features, which can describe the dynamic changes of crop polarimetric information. In this paper, we propose a tensor form to comprehensively describe the information of time-series PolSAR data, including spatial context information, polarimetric scattering information, and temporal context information. And we define a novel similarity value for the tensors (TSV), which can simultaneously consider distance and shape similarity of tensors. Then, we construct a tensor-based graph representation to capture the global similarity information of time-series PolSAR data. Finally, we propose a tensor-based graph convolutional network (Tensor-GCN) to extract deep features of graph node tensors for crop classification. Experimental results and analysis on two time-series PolSAR data firmly demonstrate the superiority of the proposed Tensor-GCN to other state-of-the-art methods.

Convolutional neural networks (CNNs) have demonstrated impressive ability to achieve promising results in PolSAR image classification. However, the traditional CNN performs convolution on local square regions with fixed sizes. The selection of these local square regions (patches) cannot fully take advantage of the boundary information of land covers and cannot search optimal neighborhoods in the whole image. To overcome these shortcomings, we propose a superpixel-based graph convolutional network (SP-GCN) for PolSAR image classification. SP-GCN utilizes superpixels as graph nodes, which makes full use of boundary information of superpixels and significantly reduces the computational cost of GCN, making it possible to apply GCN to large-scale PolSAR image classification. To reduce the impact of superpixel scale on classification results, we further propose a multiscale superpixel-based graph convolutional network (MSSP-GCN) based on the SP-GCN. Experimental results on three PolSAR datasets firmly demonstrate the superiority of the proposed SP-GCN and MSSP-GCN to other state-of-the-art methods.

Automatic registration of optical and synthetic aperture radar (SAR) images is one of the most challenging tasks due to the influence of speckle noise and nonlinear radiation differences. In this article, we propose a compound registration method for optical and SAR images based on feature decoupling network (FDNet), which consists of a residual denoising network (RDNet) and a pseudo-Siamese fully convolutional network (PSFCN). First, we propose a fast compound matching algorithm, which can overcome the respective weaknesses of the registration accuracy and computational complexity of the feature-based and area-based methods. Specifically, FAST keypoint detection is used to generate the center of the initial template. The extraction of local feature descriptors and the matching of the initial templates are implemented by PSFCN. Second, we design an RDNet to learn the statistical model of speckle noise in SAR images and define a new loss function based on mean-square error (mse) and total variation (TV) to achieve the propagation of speckle noise. PSFCN and RDNet are used to learn deep representations of semantic and noise information, respectively. Then, the semantic and noise features are decoupled on spaced convolutional layers. Finally, the optimal matching templates are searched in a small search window around the initial matching templates. In addition, we propose a strategy for adaptively selecting the template size based on 2-D entropy, which can select the appropriate template size according to the content richness of SAR images. Registration results on a public registration dataset show that our proposed method achieves better performance than other state-of-the-art methods.