Hyperspectral Imaging Combined with Multi-Scale Feature Fusion and Improved Faster R-CNN for Automated Detection of Surface Defects on Chicken Carcasses

Chicken meat holds a dominant position in the meat consumption market, and broilers are important primary processed products. However, during the slaughtering and processing of broilers, steps such as hanging, bleeding, scalding, defeathering, neck slitting, and evisceration can easily cause skin break and skin scratch on chicken carcasses due to manual operational errors. Additionally, chicken carcasses may suffer broken wing bones from intense collisions during transportation. These issues negatively impact the quality and market acceptance of broilers in subsequent sales. This study aims to develop a stable and efficient method that leverages the unique technical advantages of hyperspectral imaging information fusion and rich dimensionality. By capturing, processing, and fusing multi-scale information, the method achieves automatic detection of defects such as skin break, skin scratch, and exposed broken bones during chicken carcass processing. The improved defect detection model for chicken carcasses shows significantly enhanced performance, with average precision (AP) exceeding 92% and recall rates above 91% for all defects. The overall mean average precision (mAP) increased from 80.2% to 93.6%. This method not only provides a reliable technical approach for detecting defects in chicken carcasses but also offers a referable solution for the visual inspection of other livestock and poultry products.