Characterizing Multispectral Images of Tumorous, Bruised, Skin-torn, and Wholesome Poultry Carcasses

Abstract

A multispectral imaging system which provides image information of poultry carcasses in the spatial and the frequency domains was developed. This article discusses the spectral image characterization of poultry carcasses for separating tumorous, bruised and skin-torn carcasses from the normal carcasses based on the gray-scale intensity, Fourier power spectrum, and fractal analyses of the spectral images. Tumorous, bruised, and other damaged carcasses were scanned by an intensified multispectral camera with various wavelength filters (542-847 nm). The results showed that the spectral images scanned at the optical wavelengths of 542 and 700 nm were useful for separating the tumorous from the normal carcasses in the spatial domain. Fractal dimension, roughness and slope were calculated from the Fourier spectra of each carcass. The mean fractal dimension of the normal carcasses scanned at 542 nm wavelength was larger than that of tumorous carcasses (P < 0.005) and skin-torn carcasses (P < 0.1). Fractal dimension of tumor (542 nm) was smaller than that of bruised carcasses of 542 nm spectral images (P < 0.01). The fractal slope mean values of the normal carcasses were larger than the tumor (P < 0.005) and the bruise (P < 0.05). Whereas, the slope of bruised carcass scanned at 542 nm was much smaller than the slope of bruise scanned at 700 nm (P < 0.005). Neural network classifier performed with 91.4% accuracy for the separation of tumorous carcasses from normals based on the spectral images scanned at both 542 and 700 nm wavelengths.