SAR image ship detection algorithm based on denoising and multi-layer attention fusion

Synthetic aperture radar (SAR) image ship detection technology is the key technology of ship surveillance and recognition. However, the presence of noise in SAR images increases the difficulty of detection. To solve these problems, a ship detection algorithm combining denoising and multi-layer attention fusion (DANet) is proposed. Firstly, a non-local mean (NLM) denoising algorithm is introduced to reduce the influence of noise in SAR images and improve image quality. Secondly, a feature enhancement module (FEM) is proposed to enhance the feature extraction capability of the backbone network. Finally, a multi-layer attention fusion module (MAFM) is designed to combine the information of shallow features with that of deep features. Experimental results on the SSDD dataset show that DANet achieves 97.3% on the AP50 index, a 1.5% improvement over the baseline model.