Intensified Multispectral Imaging System for Poultry Carcass Inspection

A multispectral imaging system possessing six optical filters at 542, 570, 641, 700, 720, and 847 nm was demonstrated feasible for discriminating normal poultry carcasses from abnormal carcasses. Gray level image intensities of selected wavelengths were used to classify normal carcasses from abnormal carcasses. Visible wavelengths, 542, 570, and 641 nm, were useful for separation of normal carcasses from septicemic and cadaver carcasses. The accuracy was 89.3% for separation of normal carcasses from abnormal carcasses. The accuracy for separation of the normal carcasses from the septicemic was 89.3%, and from cadaver, 96.5%. Among the abnormal carcasses, the accuracy of classification of the septicemic carcasses was 90.0% and of the cadaver carcasses, 96.6%. The accuracies of classification for poultry carcasses into three classes using a neural networks classifier were 92.9% for normal, 83.4% for septicemic, and 96.6% for cadaver carcasses.