Lightweight Ship Object Detection Algorithm in SAR Images based on Deep Learning

Due to the small size of detection of targets and the complexity of the background in SAR images, the model has problems such as excessively complex models and low detection accuracy. Therefore, this paper presents the YOLOv8-AREB algorithm. This refined algorithm commences by eliminating the layer dedicated to large target detection from SAR imagery, streamlining the network architecture. In the main network, it uses RepConv and EMA attention mechanism to improve detection accuracy by multi-branch reparameterization and aggregation of cross-spatial dimension information. Additionally, in feature extraction stage, it incorporates BoTnet backbone to improve effectiveness of feature extraction and while enhancing overall model performance. During training, the Focal_EIoU loss is employed to enhance target detection accuracy and improve the ability to generalize. Experiments illustrate that the YOLOv8-AREB algorithm offers advantages such as high detection accuracy, low computational complexity, and minimal parameter requirements.