LEARNING VECTOR QUANTIZATION FOR COLOR CLASSIFICATION OF DISEASED AIR SACS IN CHICKEN CARCASSES

Abstract

The variation in color features observed during the evolution of air–sacculitis in chicken carcasses is exploited to classify the disease using digital imaging and neural networks. For the experiments, air–sacculitis was induced by secondary infected of E. coli via direct inoculation of challenge bacteria. Mild and severely infected samples were obtained and imaged. For the supervised classification, a knowledge base set of normalized RGB values, corresponding to negative, mild, and severely infected air sac images, was obtained. Statistical data exploration indicated no significant difference between the color features of mild and severely infected sacs, but a significant difference was found between infected and negative tissues. A neural network using the learning vector quantization algorithm classified the data in infected and negative categories. Resubstitution and hold–out errors were calculated, giving an overall accuracy in the classification of 96.7%. Each poultry carcass sold in the U.S. must be visually inspected for its wholesomeness by a USDA inspector, with air–sacculitis being the major cause of condemnation in poultry processing plants. The method presented here has the potential for integration in a computer–assisted inspection of wholesomeness in poultry processing lines.