Detection of Atlantic salmon residues based on computer vision

Atlantic salmon is an important aquaculture product. The mixture residue problem in salmon may affect food safety and quality issues. Traditional residue detection methods require the use of large or specific instruments, so a quick, low-cost, and real-time detection of residue is needed. To solve this problem, we proposed a YOLOv5n-se model, introducing SE attention mechanism into YOLOv5n. We also trained other object detection models, YOLOv5n, YOLOv5s, YOLOv5m, YOLOv5l, YOLOv8n, and Faster RCNN as the comparison algorithms. The results show that the improved model YOLOv5n-se has the best F1 score of 0.842 and the highest mean average precision (mAP50) of 0.865 which solves the problem of identifying reflections as scales and avoids mis-detect fat as bone, performs well in both quantitative and qualitative. The weight size of trained YOLOv5n-se mode only 3.75 MB, can realize salmon residue detection in a quick and low-cost way.