Real-time detection of dead fish for unmanned aquaculture by yolov8-based UAV

Abstract

As the importance of aquaculture increases, the scale and benefits of aquaculture are continuously emphasized. To achieve stable, efficient and high-precision detection instead of manual labor, we have adopted a UAV combined with artificial intelligence approach by improving the YOLOv8 model, and realizes the real-time detection of dead fish in the aquaculture by transmitting the frames captured by the UAV back to the server for computation in real time. The experimental results show that the improved model achieves an accuracy of 95.6 %, a recall of 95.2 %, a mAP 50 of 98.1 %, and a MAP50–95 of 67.5 % on the dataset constructed in this paper, which are 5.4 %, 22.8 %, 12.6 %, and 10.52 % higher than the original model, respectively. As a result, the accuracy of recognizing small dead fish targets in the view of UAV aerial photography has been greatly improved, and can be applied to the detection under different water surface shooting conditions. In addition, the UAV detection method proposed in this study can be combined with other schemes and applied to unmanned management of aquaculture and fisheries to detect problems in aquaculture in a timely and accurate manner.