HA-Net: a SAR image ship detector based on hybrid attention

Synthetic aperture radar (SAR) ship image recognition technology is essential for monitoring and identifying marine vessels. However, the existing ship detection methods are not ideal when dealing with multi-scale ships under complex backgrounds. This article proposes a SAR ship detection network (HA-Net) based on hybrid attention to solve the problem of ship detection in complex backgrounds. Firstly, this study designs the hybrid attention feature enhancement module (HAFE) and hybrid attention feature fusion module (HAFF). The HAFE and HAFF based on hybrid attention can improve the network's feature extraction and feature fusion ability. Secondly, this study presents the enhanced spatial pyramid pooling (ESPP) module to enhance the model's ability to recognize multi-scale targets by combining contextual information and capturing global dependencies. The experiments on SSDD and HRSID datasets indicate that the mAP50 of HA-Net is 96.2% and 88.2%, respectively, which is 1.6% and 2.0% better than YOLOv5. Furthermore, a comparison with other well-known detection algorithms on the ship dataset shows that HA-Net has superior detection performance. Additionally, HA-Net's FPS reaches 87.6, which can satisfy the demand for real-time ship detection.