DRA-Net: Dilated Residual Attention Network for Ship Detection in Complex Maritime Scenes

Complex coastal scenes in synthetic aperture radar ocean imagery have diverse feature types and ship targets mixed in the background, which leads to the targets being easily interfered by the coastal background clutter, so this paper proposes an algorithm based on dilated residual attention for ship target detection in complex scenes. Firstly, the dilated residual attention module is proposed to extract multi-scale features and enhance the detection ability of small targets while retaining contextual information by using cavity convolution with different parameters; secondly, a small target detection head is added and an improved decoupled detection head is designed in combination with the dilated residual attention module to improve the detection ability of small ship targets. The experimental results prove the effectiveness of the proposed method, which is tested on the HRSID dataset, and the proposed algorithm improves the mAP50 and mAP95 by 2.25% and 6.21%, respectively, compared with the benchmark model YOLOv8, and the proposed algorithm improves the mAP50 and mAP95 by 6.53% and 12.05%, respectively, in the complex littoral scenario of this dataset.