A Novel Lightweight Underwater Dense Small Object Detection Algorithm Based on YOLOv11

The underwater environment is complex and the targets to be detected are small and clustered. Meanwhile the features required to be extracted are numerous, which causes high computational complexity, and low-frequency features are easily lost. In view of this, a detection algorithm called WLF-YOLO is proposed to address the above problems, which is an improved version based on the YOLOv11 algorithm. Firstly, a C3k2-WT module is proposed to improve the extraction of low-frequency features with varying receptive fields. Then, the SPPF-LSKA module and the C2PSA-LKA module are proposed to reduce computation and enhance feature extraction, which improves the ability to detect small targets. Finally, the feature enhancement gating module (FEGM) is proposed to refine multi-scale features and detect densely object. The ablation experiments validate the effectiveness of the C3k2-WT, SPPF-LSKA, C2PSA-LKA and FEGM modules. The comparison experiments on the URPC dataset show that the AP50 of the WLF-YOLO algorithm (76.1%) is higher than the original YOLOv11 (71.7%).