YOLOv5-Based Improved Model for River Fish Detection

The models of the YOLO series are of crucial importance in the field of computer vision. YOLOv5 outperforms previous versions by being more powerful, precise, and faster, and has immense utility not only in object detection but also in other areas such as image identification. YOLOv5 outperforms previous versions as it is more potent, precise, and quicker. Nonetheless, there is still more room for development to achieve more efficient results in practical applications. This paper presents an improved version called YOLOv5-CAWIOU, which relies on the CA (Coordinate Attention) technique. In addition, CIOU has been improved. In this paper, no pre-training files are used to compare the experimental results better. Instead, YOLOv5 is optimized to improve the accuracy of model detection, robustness, and performance. The ‘Rivr Fish.v10i.YOLOv5pytorch’ dataset is employed for testing the models. The recall rate improved by 16.7%. Accuracy decreased by only 5.6%, map_0.5 improved by 6.1% and map_0.5:0.95 improved by 4.8%. Additionally, the model also has better robustness.