Computer vision-guided automatic cutting system for chicken parts

To reduce labor costs and enhance the efficiency and accuracy of poultry slaughtering andprocessing, a comprehensive solution was proposed. This system leveragesYOLOV4-CSPDarknet53 and U-Net models to recognize and localize chicken parts, respectively.The chicken parts dataset used in this study achieved mAP (mean Average Precision) of 98.86%with the YOLOv4-CSPDarknet-53 model, and the inference speed for a single image was 22.2ms. Additionally, the performance of the chicken parts image segmentation model based on U-Net yielded an MIOU (Mean Intersection over Union) of 81.03% and a pixel accuracy of 97%.Toaid in planning, a Gaussian process regression model was employed to predict the cuttingdistance of the robotic arm for chicken parts. The coefficient of determination (R) and MAE (Mean Absolute Error) for thepredicted cutting distances Y1 (for the chicken middle wing) and Y2 (for the chicken wings)based on the Gaussian process regression model were 0.69,0.35 and 0.75,0.28,respectively.The “Huichuan” robotic arm and milling cutter worked together to complete the trajectoryplanning and cutting operations of the chicken parts. The final average residual rate of themilling cutter cutting chicken legs was 11.83%, and the average residual rate of cutting chickenwings was 11.68%.