A Few Samples Underwater Fish Tracking Method Based on Semi-supervised and Attention Mechanism

Fish behavior can be used as an important indicator of aquaculture water quality, its behavior can timely and accurately reflect the quality of water. Therefore, tracking fish behavior is of great significance. However, there is currently a lack of public fish tracking datasets, and the production of datasets requires a lot of labor costs. To address this issue, we propose a few samples underwater fish tracking method based on semi-supervised and attention mechanisms. First, we use Yolov4-tiny as our base model and add the CBAM attention mechanism to it. Second, we train our model using the self-train method. Finally, we combine the improved object detector with the Sort tracker to track the underwater fish. The experimental results show that the tracking method proposed in this paper has better performance in tracking accuracy and tracking efficiency.