Deep learning has made rapid progress in the field of synthetic aperture radar (SAR) detection. However, SAR images themselves have limited information, and a general detection network that is too wide and too deep can result in computational complexity and memory waste. Therefore, we design a lightweight network for multi-class SAR detection based on adaptive scale distribution attention. Firstly, a novel backbone is designed from the perspective of lightweight model, using deep separable convolution to generate high-quality feature maps of protruding targets, and applying channel shuffle to improve training and detection efficiency. Secondly, a lightweight adaptive scale distribution attention is proposed, which can adaptively obtain the scattering information of multi-scale targets, aggregate the position and contour features of the targets, and improve the detection accuracy of multi-class targets. Finally, anchor-free detection head is applied to improve the generalization ability and robustness of the model. MLSDNet achieve a high mean average precision (mAP) of 92.99% on the newly released multi-class SAR target datasets (MSAR-1.0) with only 1.42G FLOPs and 928.25K Params. The mAP on SAR ship datasets such as SSDD and HRSID reached 99.1% and 94.7%, respectively. The mAP on the latest SAR aircraft dataset reached 97.7%, demonstrating its good generalization ability. Its performance has reached the state-of-the-art (SOTA).

Detecting ships in synthetic aperture radar (SAR) images is a challenging task due to various factors, such as the diverse distribution of ships and the intricate nature of SAR images. In recent years, deep learning has made excellent progress in the field of SAR interpretation. Models that focus on extracting global semantic information can effectively achieve balanced detection of multiscale SAR targets, but their computational complexity is relatively high. Models that focus on processing local information have redundant calculations and poor robustness, but are prone to mistaking the background information of SAR images for targets. To address the above issues, we propose a real-time SAR ship detector based on global semantic information enhancement and local gradient information mining. The lightweight feature extraction backbone based on linear computing is designed, with the network structure of global information augmentation encoder (GIAE), local gradient information miner (LGIM), and decoder, which can quickly perform feature extraction. GIAE enhances the expression of image content through the long-sequence modeling capability of the state-space model (SSM). LGIM uses gradient modules composed of depthwise separable convolutions to extract local information of image and utilizes directed self-attention (DSA) to mine channel context information. Global-local detector (GLDet) can complete object detection, rotated object detection, and instance segmentation tasks by transforming the detection head. Excellent performance has been achieved on the SAR ship instance segmentation datasets: SAR ship detection dataset (SSDD) and high-resolution SAR image dataset (HRSID), as well as the SAR rotated ship datasets: rotated ship detection dataset (RSDD)-SAR and SSDD+. Meanwhile, GLDet demonstrated excellent generalization performance in large-scale SAR images captured by GF-3 and Terra-SAR satellites.

Synthetic aperture radar (SAR) images of ships have complex background interference, multi-scale targets, and irregular distribution characteristics. However, existing mainstream SAR ship detection algorithms rely on manually designed hyperparameters. This results in poor robustness, which makes it difficult to effectively balance the detection accuracy and speed. To solve these problems, a novel anchor-free SAR ship detection algorithm based on self-balancing position attention is proposed. First, a lightweight feature extraction backbone (GhostVS-Net) is designed by FOCUS, Ghost, and separable convolution modules to extract feature information. This helps to extract the contour features of ships and suppress unrelated interference, making it more suitable for the scattering characteristics of SAR images. Second, a self-balancing position attention (SBPA) module is designed, which balances the local image features under multiple receptive fields, and aggregates the global position information and spatial context information. The proposed SBPA module considers the characteristics of background, scale, and distribution of SAR images to obtain greatly improved positioning accuracy. Finally, the feature pyramid network (FPN) is applied to fuse the scale context information, and further improve the accuracy of detection. Experimental results on the SSDD and HRSID datasets show that the proposed network bears feature of accurate detection and strong robustness, with the mean average precision (mAP) reaches 99.72% and 95.30%, which reveals that it outperforms all current state-of-the-art algorithms with super high performances.

With the significant advancements in deep learning technology and the substantial improvement in remote sensing image resolution, remote sensing semantic segmentation has garnered widespread attention. Synthetic aperture radar (SAR) and optical images are primary sources of remote sensing data, offering complementary information. SAR images can capture surface information even under cloud cover and at night, whereas optical images provide higher resolution in clear weather conditions. Deep learning-based feature fusion methods can effectively integrate multi-source information to obtain more comprehensive surface data. However, there are significant spatiotemporal differences in multi-source information, making it challenging to select and extract the most discriminative features for segmentation tasks. To address this, we propose a lightweight and efficient fusion semantic segmentation network, SOLSTM, which mixes SAR and optical images as inputs and performs cyclic cross-fusion to establish a new network paradigm. To tackle multi-source data heterogeneity, we introduce SAR-OPT matching attention, which aggregates multi-source image features by adaptively adjusting fusion weights, thereby achieving comprehensive perception of feature channels and contextual information. Additionally, to mitigate the high computational complexity of processing multidimensional data, we introduce the mLSTM block, which employs linear operations to mine global contextual information in fused images, thus reducing computational complexity and enhancing image segmentation performance. Experiments on the WHU-OPT-SAR dataset show that SOLSTM has excellent performance, achieving up to 52.9 mIoU and outperforming single source image segmentation, verifying the effective fusion of OPT-SAR.

Intelligent ship detection based on synthetic aperture radar (SAR) is vital in maritime situational awareness. Deep learning methods have great advantages in SAR ship detection. However, the methods do not strike a balance between lightweight and accuracy. In this article, we propose an end-to-end lightweight SAR target detection algorithm, multi-level Laplacian pyramid denoising network (LPDNet). Firstly, an intelligent denoising method based on the multi-level Laplacian transform is proposed. Through Convolutional Neural Network (CNN)-based threshold suppression, the denoising becomes adaptive to every SAR image via back-propagation and makes the denoising processing supervised. Secondly, channel modeling is proposed to combine the spatial domain and frequency domain information. Multi-dimensional information enhances the detection effect. Thirdly, the Convolutional Block Attention Module (CBAM) is introduced into the feature fusion module of the basic framework (Yolox-tiny) so that different weights are given to each pixel of the feature map to highlight the effective features. Experiments on SSDD and AIR SARShip-1.0 demonstrate that the proposed method achieves 97.14% AP with a speed of 24.68FPS and 92.19% AP with a speed of 23.42FPS, respectively, with only 5.1 M parameters, which verifies the accuracy, efficiency, and lightweight of the proposed method.

Synthetic Aperture Radar (SAR) encounters distinct challenges in airborne surveillance (dynamic scene variations, target edge blurring) and spaceborne observation (large-scale analysis, high-resolution processing). Both traditional methods and contemporary deep learning-based solutions exhibit limitations: inadequate dynamic target adaptability, weak small-target detection, and redundant recognition in large-scale scenarios, stemming from challenges like target ambiguity, occlusion, and inter-class similarity. To address these challenges, we propose the Regional Spatial-Channel Information Weighted Attention Network (RSCIWANet). The innovations encompass: 1) Regional Spatial Channel Attention (RSCIWA) integrates regional weighting in spatial attention to amplify key positional features while suppressing speckle noise and edge weak samples. Channel self-attention enhances cross-regional interactions to capture target-environment scattering correlations. 2) Boundary-Aware Loss (BA_LOSS) employs edge overlapping penalties to improve localization of fuzzy shadow edges, with adaptive weighting to amplify small-target gradient contributions during backpropagation. 3) Context-Preserving Sliding window detection strategy for large-scale images, which can carry out comprehensive and robust detection. Experimental results demonstrate state-of-the-art performance, with the mAP⁵⁰ of 99.35% on Sandia National Laboratories (SNL) video SAR dataset, 97.50% on MSAR-1.0 dataset, and superior large-scale detection capability on MSAR-1.0 and LS-SSDDD datasets.