Mechanisms of Machine Vision Feature Recognition and Quality Prediction Models in Intelligent Production Line for Broiler Carcasses

With global broiler production reaching 103 million tons in 2024—a 1.5% increase over 2023—the poultry industry continues to grow rapidly. However, traditional broiler segmentation methods struggle to meet modern demands for speed, precision, and adaptability. First, this study proposes an improved lightweight image segmentation algorithm based on YOLOv8-seg and integrates the Segment Anything Model (SAM) for semi-automatic annotation, achieving precise mask segmentation of broiler parts. Subsequently, Key geometric features (e.g., area, perimeter, axes) were extracted using image processing techniques, with enhancements from HSV color transformation, convex hull optimization, and ellipse fitting. Furthermore, Image calibration was applied to convert pixel data to physical dimensions, enabling real-sample validation. Using these features, multiple regression models—including CNNs—were developed for carcass quality prediction. Finally, by analyzing the broiler segmentation process, machine vision techniques were effectively integrated with quality grading algorithms and applied to intelligent broiler segmentation production lines, providing technical support for the intelligent and efficient processing of poultry products. The improved YOLOv8-seg model achieved [email protected]:box scores of 99.2% and 99.4%, and the CNN model achieved R2 values of 0.974 (training) and 0.953 (validation). Compared to traditional systems, the intelligent broiler cutting line reduced failure rates by 11.38% and improved operational efficiency by over 3%, offering a reliable solution for automated poultry processing.