An Underwater Object Detection Algorithm Based on Improved YOLOv5s

To tackle the challenges of object detection in complex underwater environments, we propose an underwater object detection algorithm based on the improved YOLOv5s network. Initially, to address the non-rigid deformation and multi-scale characteristics of underwater targets, a D3 module is proposed to enhance the backbone network’s capability for feature extraction, and Space-to-Depth (SPD) convolution is incorporated to reduce the depletion of detailed information in target images. Additionally, to boost the detection precision of small targets, we deploy a object detection head specifically designed for small objects and incorporate the concept of the Efficient Channel Attention (ECA) mechanism, then further integrate and utilize shallow feature information. Experiments demonstrate that the proposed underwater object detection algorithm achieves a mean Average Precision (mAP) of 86.7% on the DUO underwater target dataset. In comparison to the baseline algorithm YOLOv5s, the mAP of our algorithm is improved by 2.5%.