A study on the application of improved YOLOv5 model for fish target detection

Fish target detection is of great significance for research in production automation and so on. In order to quickly and accurately get the location of fish targets and their categories, this paper proposes a fish target detection method with improved YOLOv5 model. The experimental results show that the mAP of the improved YOLOv5 model is significantly improved over the original model, in which the WIoU loss function is introduced to improve the accuracy of the regression frames, and the accuracy on the self-constructed dataset is 92.32%, which is an improvement of 5.80 percentage points compared with the baseline model. The YOLOv5-based algorithm enables fast and accurate identification of fish localization and species, and meets real-time requirements.