ALF-YOLO: Enhanced YOLOv8 based on multiscale attention feature fusion for ship detection

Ship detection plays a crucial role in ensuring maritime transportation and navigation safety. However, accurately detecting multiscale ships remains a challenge due to the diversity of ship categories and locations, as well as interference from complex environments. Object detectors based on the You Only Look Once (YOLO) framework have demonstrated remarkable accuracy in automatic ship detection. In this paper, we integrate the Asymptotic Feature Pyramid Network (AFPN), Large Selective Kernel Attention Mechanism (LSK), and the fourth detection head into YOLOv8, developing a novel ALF-YOLO architecture. ALF-YOLO utilizes AFPN to enrich feature representation by integrating multiscale high-level semantic features and spatial details. It also incorporates a large selective kernel attention mechanism that dynamically adjusts its large spatial receptive field to focus more on crucial ship features, eliminating interference from complex environmental factors to enhance discriminative feature representations of ships. Additionally, we investigate the impact of different attention mechanisms on ship detection accuracy. Experimental results indicate that by integrating the outputs of several modules, our proposed ALF-YOLO model improves the classification and localization capability of targets at each stage. Compared to YOLOv8, ALF-YOLO achieved a relative increase of 0.41% and 0.43% in [email protected] on the Seaships and McShips datasets, respectively. Across different evaluation criteria, the overall performance of the ALF-YOLO method surpasses existing ship detection methods.