A Novel Three-Stage PolSAR Ship Detection Framework based on Deep Learning

Polarimetric Synthetic Aperture Radar (PolSAR) image ship detection is one of the crucial applications of PolSAR systems and holds significant research value. However, current deep learning-based polarimetric SAR image ship detection methods face challenges such as difficult post-processing and slow detection speed. To solve the above problems, we propose a novel Three-Stage Polarimetric SAR Ship Detection Framework based on deep learning. The main concept is to use a sliding window to extract image slice samples, then employ the Ship Coarse Location Module (SCLM) to identify slice samples containing ships. Next, the Polarization Feature Enhancement Module (PFEM) enhances the features of these slice samples to reduce the feature distance of pixels of the same category and increase the feature distance of pixels of different categories. Afterward, the enhanced slice samples are fed into the KAN Vision Transformer Detection Module (KVITDM) for patch-level detection. Finally, decision fusion is applied to combine the detection results generated from different initial sliding window positions. Experimental results show that the proposed method has good performance in different scenes and outperforms traditional and existing deep learning-based polarimetric SAR ship detection methods.