SAR SHIP DETECTION METHOD BASED ON CNN-TRANSFORMER HYBRID WITH DYNAMIC FREQUENCY DOMAIN ATTENTION

Synthetic Aperture Radar (SAR) ship detection has attracted significant attention due to its important value in various fields such as maritime management, traffic control, and marine protection. The SAR ship detection based on CNN and Transformer has many applications. However, due to the limitation of traditional convolution operations in adapting to input transformations because of their static nature, as well as the deficiencies of Transformers in capturing local details, This paper proposes a ship detection method based on CNN-Transformer hybrid with dynamic frequency-domain attention. CNN and Transformer network branches are utilized to extract local features and global information from the image, respectively, and an interactive fusion is achieved through a label-mixing module to enable more precise object detection. With the effective guidance of the dynamic frequency-domain attention module, the model is better able to distinguish subtle differences between target features and background noise, providing strong technical support for ship target detection tasks in complex environments. Extensive experiments conducted on multiple datasets validate the effectiveness of the proposed method in this paper.