Low visibility underwater biological target detection based on YOLOV5s

Underwater robots play an important role in aquatic organisms’ fishing , but the low visibility may lead to poor recognition of marine organisms’ fishing. To improve this issue, we propose an improved YOLOv5s model. Given the problem of excessive invalid information in underwater images, this method integrates a Coordinate Attention (CA) mechanism into the feature extraction network, enhances the feature extraction capability of the network, reduces feature redundancy , and helps the model locate underwater organism targets more quickly and accurately. At the same time, the Bi-directional Feature Pyramid Network (BiFPN) is used to replace the Path Aggregation Network (PAN) structure, and the features of different scales are fused quickly. Finally, the Structured Intersection over Union (SIoU) loss function is introduced to replace the Complete IoU (CIoU) loss function as a new boundary frame loss function, which improves the convergence speed and accuracy of the model. The experimental results show that compared with other algorithms, the improved YOLOv5s algorithm has more advantages in accuracy and detection speed.