Visible/Near-infrared Reflectance and Interactance Spectroscopy for Detection of Abnormal Poultry Carcasses

Abstract

The results of the development of a nondestructive and noninvasive technique for the detection of abnormal poultry carcasses using a diode array spectrophotometer system were presented. Visible/near-infrared reflectance and interactance spectra (500 to 1113 nm) for both sides of the breast of normal, septicemic, and cadaver poultry carcasses were obtained using a fiber optic probe with the spectrophotometer. Optimal wavelengths for correlating the spectral reflectance and interactance and second difference with the condition of the poultry carcasses were obtained. In general, differentiating between the normal and cadaver carcasses was more accurate than differentiating between normal and septicemic and between septicemic and cadaver. It was concluded that this technique can be used to separate normal carcasses from the abnormal (septicemic and cadaver) ones with good accuracy. The model based on second differences of interactance spectra, which requires 24 wavelength readings, yielded a classification accuracy of 93.3% for normal carcasses and 96.2% for the abnormal carcasses. Alternative decision criteria could be chosen so that none of abnormal carcasses were misclassified as normal. However, for a sensor limited to four wavelength readings, the model based on the reflectances should be used. The reflectance model required four wavelength readings and provided a classification accuracy of 94.4% for normal carcasses and 93.1% for abnormal carcasses.