Enhancing Fish Detection and Classification in Sonar Images Through Deep Learning

To ensure the preservation of a robust ecosystem and healthy fish populations, it is necessary that we carefully monitor the number and presence of different fish species. In this article, we fine-tuned and tested an automated real-time state-of-the-art deep learning model YOLOV9. We showed that an adaptation of the widely used YOLO deep learning model can accurately detect and classify eight species of fish from a public DIDSON imaging sonar dataset captured from the Ocqueoc River in Michigan, USA. Our results demonstrate that deep learning models can be used to detect and classify fish using high resolution imaging sonar. We obtained a high score with a mean Average Precision of 0.894 for YOLOv9-e and 0.871 for YOLOv8xl (mAP:50).