YUWNet: improved underwater target detection algorithm of YOLO11n

Underwater target detection technology is widely applied in fields such as ecological protection, economic development, and scientific research, promotion of sustainable development and utilization of marine resources. To address issues such as poor image quality, target diversity, occlusion, and limited resources of underwater equipment, an improved YOLOv11n underwater target detection algorithm, YUWNet, is proposed. Firstly, the backbone network is improved by using DualConv to reduce the probability of missed and false detections caused by target occlusion and complex backgrounds in underwater images. Then, an improved Transformer structure is integrated into the part of the feature map with smaller dimensions to mitigate the impact of large-scale global information issues such as blurring effects on model performance. At the same time, an efficient downsampling module, ADown, is introduced to reduce the computational load. To address the problems of precise positioning, detail recognition, small object detection, and object boundary separation of underwater organisms, the advantages of edge detection technology and the Bifpn network are combined to design a new SBifpn neck structure. Finally, a lightweight detection head LSCD is introduced, which reduces the parameters and FLOPs while maintaining good detection performance. On the URPC2021 dataset, the experimental results show that the mAP50 and mAP50-95 of YUWNet are 2.0% and 2.1% higher than that of YOLO11n, the parameters is decreased by 46.2% and the FLOPs is reduced by 7.9%.