Hyperspectral Image Classification for Fecal and Ingesta Identification by Spectral Angle Mapper

Abstract

This paper presents the performance of spectral angle mapper (SAM) supervised classification method for hyperspectral poultry imagery to classify fecal and ingesta contaminants on the surface of broiler carcasses. Spatially averaged spectra of three different feces from the duodenum, ceca, colon, and ingesta of corn/soybean diet were used for classification data. SAM classifier using reflectance of hyperspectral data with 512 narrow bands from 400 to 900 nm was able to classify three different feces and ingesta on the surface of poultry carcasses. Based on the comparison with ground truth region of interest, both classification accuracy and kappa coefficient increased when spectral angle increased. The overall mean accuracy and corresponding mean kappa coefficient to classify fecal and ingesta contaminants were 90.13% (standard deviation = 5.40%) and 0.8841 (standard deviation = 0.0629) when a spectral angle of 0.3 radians was used as a threshold.