Abstract: This paper presents a method for the identification of diluted contaminants on the surface of chicken carcasses based on successive projections algorithm (SPA)-multiple linear regression (MLR)-receiver operating characteristic (ROC) classifier. Firstly, a total of 20 images of carcasses with diluted contaminants were acquired by hyperspectral imaging system, and 10 characteristic bands were extracted from the 1 232 bands by SPA. Then the MLR method was used to construct a regression model between the discriminant function and the characteristic spectral bands. Finally, the optimal classification threshold with high true positive rate (TPR) and low false positive rate (FPR) was determined by ROC analysis. Thus, the SPA-MLR-ROC classifier allowed the identification of the diluted contaminants. The results showed that the TPR of the SPA-MLR-ROC classifier was 98.08% and the FPR was only 0.39%. The detection accuracy was higher than that of the band ratio algorithm and the dual-band algorithm. Hence, the SPA-MLR-ROC classifier exhibited good performance for the detection of diluted contaminants on the surface of chicken carcass. However, because of the limited number of samples, further study using more samples is needed to verify the stability and feasibility of this method.

Key words: successive projections algorithm-multiple linear regression-receiver operating characteristic analysis, classifier, contaminants, hyperspectral imaging, chicken carcass