Abstract:The fundamental challenge in SAR target detection lies in developing discriminative, efficient, and robust representations of target characteristics within intricate non-cooperative environments. However, accurate target detection is impeded by factors including the sparse distribution and discrete features of the targets, as well as complex background interference. In this study, we propose a \textbf{Ma}mba \textbf{Di}ffusion \textbf{Net}work (MaDiNet) for SAR target detection. Specifically, MaDiNet conceptualizes SAR target detection as the task of generating the position (center coordinates) and size (width and height) of the bounding boxes in the image space. Furthermore, we design a MambaSAR module to capture intricate spatial structural information of targets and enhance the capability of the model to differentiate between targets and complex backgrounds. The experimental results on extensive SAR target detection datasets achieve SOTA, proving the effectiveness of the proposed network. Code is available at \href{<a class="link-external link-https" href="https://github.com/JoyeZLearning/MaDiNet" rel="external noopener nofollow">this https URL</a>}{<a class="link-external link-https" href="https://github.com/JoyeZLearning/MaDiNet" rel="external noopener nofollow">this https URL</a>}.

What problem does this paper attempt to address?

The problems that this paper attempts to solve mainly focus on several key challenges in Synthetic Aperture Radar (SAR) target detection: 1. **Targets Composed of Discrete Echo Points**: - Targets (such as airplanes, ships, etc.) in SAR images are presented as sets of discrete echo points, lacking the geometric shapes, contours, and texture features commonly seen in natural images. This characteristic makes it difficult for traditional anchor - based methods to accurately locate and distinguish targets, and it is prone to over - detection and false positives. 2. **Diversity in Target Sizes and Sparse Distribution**: - SAR images usually cover large scenes, where targets are relatively sparse and vary greatly in size. Traditional anchor - based detection methods, due to using fixed, uniformly - distributed proposal boxes, are likely to lead to missed detections and false detections of irregular - shaped targets, while also increasing computational complexity and resource consumption. 3. **Complex Background Interference**: - The scattering characteristics of background structures and metal objects in SAR images are highly similar to those of targets, which makes feature - based discrimination difficult. Therefore, how to effectively use context information to reduce background interference and enhance target feature representation is an important issue. To solve the above problems, this paper proposes the Mamba Diffusion Network (MaDiNet), and its core contributions are as follows: - **Introducing the Diffusion Model for Target Detection**: The SAR target detection task is regarded as the process of generating real target boxes from noise boxes. The diffusion model is used to directly generate the positions (center coordinates) and sizes (width and height) of target bounding boxes in the image space. This method significantly improves the inference accuracy of the detector and is the first work to combine the diffusion model with Mamba for SAR target detection. - **Designing the MambaSAR Module**: In order to capture the rich spatial structure information of targets, especially their position information in the image, the MambaSAR module is designed, which enhances the model's ability to distinguish targets from complex background interference. - **Extensive Experimental Verification**: A comprehensive analysis of MaDiNet was carried out on three representative datasets, covering more than 20 mainstream detection methods (including anchor - based and anchor - free methods). Quantitative and qualitative experimental results show that MaDiNet outperforms all comparison models in detection performance. In conclusion, this paper aims to solve the key challenges in SAR target detection through the innovative diffusion model and the specially - designed MambaSAR module, thereby achieving more accurate and efficient target recognition.

MaDiNet: Mamba Diffusion Network for SAR Target Detection