Discriminant analysis of dual-wavelength spectral images for classifying poultry carcasses

An analysis of texture features, based on co-occurrence matrices (COMs), was conducted to determine the performance of dual-wavelength imaging for discriminating unwholesome poultry carcasses from wholesome carcasses. The variance, sum average, sum variance, and sum entropy of COMs were the most significant texture features (P < 0.005) for identifying unwholesome poultry carcasses. However, the feature values of angular second moment, variance, sum average, sum variance, and sum entropy did not vary with the COM parameters, distance and direction. The characteristics of variance and sum variance texture features varied with the wavelength of spectral images and with condemnation of poultry carcasses, as well. The sum variance of wholesome carcasses was higher (P < 0.005) than unwholesome carcasses for spectral images at 542 nm. For 542 and 700 nm images, linear discriminant models were able to identify unwholesome carcasses with a classification accuracy of 91.4%. However, a single linear discriminant model was not acceptable for identifying three different types of carcasses (wholesome, septicemic and cadaver), because of extreme inaccuracy for septicemic carcasses. In this case, the classifier that demonstrated the highest accuracy was 89.6% accurate at 542 nm. Thus, a dual-wavelength imaging system with optical filters of 542 and 700 nm wavelengths appears promising for detecting unwholesome poultry carcasses. Published by Elsevier Science B.V.