The convolutional neural network (CNN) is widely used in synthetic aperture radar (SAR) target recognition, but conventional CNN mainly adopts a single-scale convolutional kernel, resulting in losing part of the feature information of targets and does not pay enough attention to significant features. On the other hand, conventional CNN approaches only assign fine-class labels to SAR targets, ignoring the high-level semantics information of similar categories, which reduces the feature differences between categories and the generalization ability of the model. Therefore, this article proposes a multiscale attention super-class CNN (MSA-SCNN) for SAR target classification. First, MSA-SCNN combines multiscale feature fusion with the attention module to improve the integrity of SAR target feature representation. The attention module includes channel and spatial attention modules, which realize the weighted enhancement of different scale features. Additionally, MSA-SCNN introduces super-class labels to increase the feature difference between categories. The classification stage consists of a fine-class branch and a super-class branch, and the features trained on the super-class branch are fused to the fine-class branch to improve the network's fine classification ability. Experiments on the moving and stationary target acquisition and recognition dataset and the FUSAR-Ship dataset show that the proposed MSA-SCNN outperforms many current existing state-of-the-art methods.

Due to the significant discrepancies in the distribution of ships in nearshore and offshore areas, the wide range of their size, and the randomness of target orientation in the sea, traditional detection models in the field of computer vision struggle to achieve performance in SAR image ship target detection comparable to that in optical image detection. This paper proposes an oriented ship target detection model based on the YOLO11 algorithm, Neural Swin Transformer-YOLO11 (NST-YOLO11). The proposed model integrates an improved Swin Transformer module called Neural Swin-T and a Cross-Stage connected Spatial Pyramid Pooling-Fast (CS-SPPF) module. By introducing a spatial/channel unified attention mechanism with neuron suppression in the spatial domain, the information redundancy generated by the local window self-attention module in the Swin Transformer Block is cut off. Furthermore, the idea of cross-stage partial (CSP) connections is applied to the fast spatial pyramid pooling (SPPF) module, effectively enhancing the ability to retain information in multi-scale feature extraction. Experiments conducted on the Rotated Ship Detection Dataset in SAR Images (RSDD-SAR) and the SAR Ship Detection Dataset (SSDD+) and comparisons with other oriented detection models demonstrate that the proposed NST-YOLO11 achieves state-of-the-art detection performance, demonstrate outstanding generalization ability and robustness of the proposed model.

Synthetic aperture radar (SAR) multi-view observations can acquire omnidirectional scattering characteristics of targets, providing richer information for target detection and recognition. Deep learning has been widely applied to SAR image interpretation in recent years. However, the sensitivity of SAR images to imaging parameters and the limited training samples pose challenges in applying deep learning to SAR target detection. Therefore, this paper first designs a SAR omnidirectional subaperture division data enhancement method with different aspect accumulation angles, multi-looks numbers, and sub-aperture overlap angles based on SAR imaging parameters and the original echo. Meanwhile, an omnidirectional SAR target detection dataset called SAR-Vehicle-Det is constructed using this sub-aperture division method to evaluate the model performance. Then, this paper proposes a novel network ASC-U²Det based on attributed scattering center (ASC) guidance for SAR target detection. The prediction is divided into ASC reconstruction image branch and target detection branch. The reconstruction branch uses ASC reconstruction mask maps to guide the network to extract the SAR target scattering features at different aspect angles, suppress the background clutter interference, and improve the detection accuracy of the target detection branch. Finally, this paper also evaluates the impact of imaging parameters with different subaperture accumulation angles and multi-look numbers on the network detection performance. Experiments on the SAR-Vehicle-Det and miniSAR datasets show that the proposed ASC-U²Det outperforms many existing target detection algorithms.

In recent years, deep learning technology has made significant progress in the field of target detection and recognition in Synthetic Aperture Radar (SAR) images, yet its application still faces complex challenges. Despite deep neural network models achieving recognition rates exceeding 99% under standard operating conditions on the MSTAR dataset, the unique imaging mechanisms of SAR, its background dependency, variations in imaging parameters, and diversity in preprocessing lead to highly variable image statistical characteristics, thereby affecting the performance of deep learning models. Dataset bias, particularly the bias induced by different SAR image quantization methods, is one of the key factors impacting the generalization capability of models. This paper delves into the impact of quantization methods on neural network performance and analyzes strategies to overcome dataset bias to enhance the stability and generalization capability of SAR image recognition. Experimental results indicate that models trained with adaptive quantization can learn more general features; linear quantization exhibits poor generalization when not enhanced, but this can be improved through data augmentation. Furthermore, pre-training and data augmentation techniques significantly enhance the classification performance of models under different quantization strategies, providing scientific evidence for optimizing SAR imaging system design and constructing reasonable datasets.