Adaptive Spatial Feature Fusion-Based SAR Ship Detection Algorithm

Ship target detection in SAR images is crucial for the national security of coastal countries worldwide. To address the issues of low detection accuracy and missed targets caused by densely packed ships in SAR images, we propose a ship target detection algorithm based on an improved YOLOv8. Firstly, we replace the YOLOv8 backbone network with HGNetV2, which enhances feature extraction capabilities while reducing the number of model parameters. Secondly, we improve the original detection head of YOLOv8 by introducing an adaptive spatial feature fusion method. Comparative experiments show that the mAP50 of the improved network model reaches 98.9% and 75.1%, representing an improvement of 1% and 1.1% over the YOLOv8s model. This is significant for the task of ship detection in SAR images.