A two-stage sparse structure representation algorithm which can preserve the manifold structure of the data is proposed for synthetic aperture radar target configuration recognition in this paper. Manifold structure of the data is preserved by two stages. In the training stage, taking advantage of both the sparse representation (SR) and manifold learning, local structure of the data is preserved in the reconstruction space, where SR-based recognition is realized. In the testing stage, two structure preserving factors based on the testing samples are embedded into the SR model to enhance structure preserving performance. The first one is constructed to preserve the local structure of the testing samples, which can guarantee the samples that are close to each other in the original space will also be close to each other in the sparse space. And the second one is established to preserve the distant structure of the testing samples, which can ensure the samples that are far from each other in the original space will also be far from each other in the sparse space. Manifold structure of the data is well captured and preserved by two stages. Experimental results on the moving and stationary target acquisition and recognition database demonstrate the effectiveness of the proposed algorithm.

Superpixel can maintain the boundary of the target and reduce the influence of speckle noise, which has been widely applied to synthetic aperture radar (SAR) image target detection. But the size of the superpixel has a great impact on the performance of superpixel-based SAR target detection algorithms. To solve this problem, we propose a multi-level ship target detection algorithm based on superpixel segmentation. Firstly, the SAR images are segmented in different levels with different superpixel sizes. Different descriptions of the SAR images are obtained in different levels. Secondly, we determine the feature of the superpixels in each level. And in order to enhance the adaptability of the proposed algorithm, we propose an adaptive distance calculation method to select the contrast superpixels in each level. Thirdly, the soft detection results are realized in each level by using the fuzzy C-means (FCM) algorithm. At last, the soft detection results obtained in different levels are fused by a new fusion strategy to achieve the final ship target detection result. The influences caused by different superxiel sizes can be effectively eased by fusion. Experiments in different SAR images have verified the effectiveness of the proposed algorithm in accurately detecting ship targets and insensitivity to the superpixel size.

Infrared (IR) search and track systems are widely applied in aerospace and defense fields. Infrared small target detection (IRSTD) in heavy clouds and chaotic terrestrial environments remains a challenging task. The semantic features of IR small targets are highly prone to vanishing with the addition of network layers. Transformer with quadratic computational complexity struggles for local feature refinement. To tackle this issue, we introduce a Mamba-driven approach dubbed Spatial- Spectral Mamba Interactive Learning (SMILE) network. Specifically, the perspective transformation structures heterogeneous backgrounds. The reconstructed data couples Mamba’s flattened multidirectional scanning mechanism. Given that small targets possess sparse and high-frequency properties, spatial Mamba and spectral Mamba collaboratively enrich the semantic features of small targets. The Dual-Path Aggregation (DPA) network is engineered to integrate the saliency and high-frequency attributes of small targets, effectively balancing detailed and contextual features without overwhelming the network. The Hybrid Representation Learning Module (HRLM) refines the local features to inscribe the intact edge structure. Both qualitative and quantitative experiments demonstrate that our proposed SMILE outperforms 14 recent benchmark algorithms on two public datasets.

Locality preserving projections (LPP) can preserve the local structure of the datasets effectively. However, it is not capable of separating the samples that are close to each other in the high-dimensional space but belong to different classes. Focusing on the problem, a class-dependent structure preserving projections (CDSPP) algorithm is proposed in this paper to realize synthetic aperture radar (SAR) target configuration recognition. The class information is embedded into the LPP model, and the similarity matrix and the difference matrix are constructed according to the class information. The similarity matrix is utilized to preserve the local structure of the samples belong to the same class, which makes the samples with the same class become more compact after feature extraction. And the difference matrix is utilized to separate the samples that are close to each other in the high-dimensional space but belong to different classes. Target aspect angle sensitivity of SAR images can be eased by using the proposed CDSPP algorithm. Experiments are conducted on the moving and stationary target acquisition and recognition database. The results verify the effectiveness of the proposed algorithm, and comparisons with other algorithms further prove its advantage.

Generative adversarial network (GAN) can generate diverse and high-resolution images for data augmentation. However, when GAN is applied to the synthetic aperture radar (SAR) dataset, the generated categories are not of the same quality. The unrealistic category will affect the performance of the subsequent automatic target recognition (ATR). To overcome the problem, we propose a reinforced constraint filtering with compensation afterwards GAN (RCFCA-GAN) algorithm to generate SAR images. The proposed algorithm includes two stages. We focus on improving the quality of easily generated categories in Stage 1. Then, we record the categories that are hard to generate and compensate by using traditional augmentation methods in Stage 2. Thus, the overall quality of the generated images is improved. We conduct experiments on the moving and stationary target acquisition and recognition (MSTAR) dataset. Recognition accuracy and Fréchet inception distance (FID) acquired by the proposed algorithm indicate its effectiveness.