Abstract
Tilapia fillet is an aquatic product of great economic value. Detection of impurities on tilapia fillet surfaces is typically performed manually or with specialized optical equipment. These residues negatively impact both the processing quality and the economic value of the product. To solve this problem, this study proposes a tilapia fillet residues detection model, the double-headed GC-YOLOv10n; the model is further lightweighted and achieves improved detection performance compared to the double-headed GC-YOLOv10n. The model demonstrates the best overall performance among many mainstream detection algorithms with a small model size (3.3 MB), a high frame rate (77FPS), and an excellent mAP (0.942). It is able to complete the task of tilapia fillet residues detection with low cost, high efficiency, and high accuracy, thus effectively improving the product quality and production efficiency of tilapia fillets.