An Improved YOLO-v4 Algorithm for Recognition and Detection of Underwater Small Targets

Since the underwater complex environmental conditions seriously affect the accuracy of underwater target recognition, in view of the low accuracy of traditional image recognition algorithms for target recognition and the lack of robustness in extracting features, an image enhancement method for small underwater targets is proposed. A scheme combined with the YOLO-v4 improved algorithm. Firstly, the K-means++ algorithm is used to improve the original input anchor box mechanism to achieve accurate anchor box selection, and the image data enhancement strategy based on geometric and photometric transformation is used to optimize the training network. Finally, the EIOU loss function is used to optimize the model performance and improve the loss function convergence speed. The experimental results show that the average accuracy rate of the improved network model reaches 98.37%, and the number of frames per second of image transmission reaches 35.69FPS, which is 4.4.% higher than the original algorithm.