Multi-object Detection in Underwater Sonar Images Based on an Improved YOLOv5 Model

In order to realize the task of underwater multi-target detection, this paper proposes an underwater multi-target detection method for sonar images based on the improved model of YOLOv5. The method introduces the NWD algorithm to alleviate class imbalance issues and utilizes the CBAM algorithm to enhance feature extraction capabilities, effectively improving the model's performance in detecting small targets in sonar images. Experimental results show that the improved YOLOv5 model significantly enhances both accuracy and mAP metrics, with accuracy increasing from 0.950 to 0.967 and [email protected] rising from 0.955 to 0.964, particularly excelling in complex scenes. Moreover, the improved model outperforms the YOLOv5s model across all categories, with performance gains ranging from 0.001 to 0.021, with the most notable improvements observed in the square cage and circle cage categories, with increases of 0.021 and 0.019, respectively. These results indicate that the improved YOLOv5 model enhances generalization ability and accuracy, better addressing the complexities and challenges of real-world scenarios.