This paper describes a transportable spectrophotometer system developed for real-time classification of poultry carcasses on-site at slaughter plants. The system measures the spectral reflectance of poultry carcasses in the visible/near-infrared regions (471 to 963.7 nm). An optimal neural network classifier for real-time classification of poultry carcasses into normal, septicemic, and cadaver classes with an average accuracy of 93% was obtained. When the classifier was used to classify the carcasses into two classes, normal and abnormal (septicemic and cadaver), the average accuracy was 97.4%. The percentages of the false positive and the false negative error rates were 2.4 and 2.9%, respectively. This paper also proposes implementing the system at the slaughter plants as a poultry carcass screening system (PCSS). Using two visible/NIR spectrophotometer systems, the PCSS tests both sides of the breast of each bird. With the PCSS, the inspection-passed-bird and inspection-rejected-bird error rates by the spectrophotometer systems would be minimal, and less than 5% of the incoming birds would require further inspection by human inspectors.
Yud-Ren Chen, Roy Winfield Huffman, Bosoon Park, and Minh Nguyen, "Transportable Spectrophotometer System for On-Line Classification of Poultry Carcasses," Appl. Spectrosc. 50, 910-916 (1996)