Prediction of pork storage time using Fourier transform near infrared spectroscopy and Adaboost‐ULDA

Abstract

AbstractPork storage time is related to its freshness. To discriminate the pork storage time rapidly, Adaboost‐ULDA algorithm was proposed to classify the Fourier transform near infrared reflectance (FT‐NIR) spectra of pork samples acquired by a Fourier transform near‐infrared spectrophotometer. Adaboost‐ULDA can not only extract discriminant information from FT‐NIR spectra but also construct a strong classifier with weak classifiers to classify spectra. Adaboost‐ULDA is a powerful classifier by combining uncorrelated linear discriminant analysis (ULDA) with Adaboost. Experimental results showed that Adaboost‐ULDA achieved the highest classification accuracy (100%) and classification accuracies were obtained as 87.9, 89.4, and 97.7% in principal component analysis (PCA) plus linear discriminant analysis (LDA), ULDA, and Adaboost‐PCA + LDA, respectively. In addition, the experiments demonstrated that the classification time of Adaboost‐ULDA was much less than that of Adaboost‐PCA + LDA. The overall results show that Adaboost‐ULDA combined with FT‐NIR is a feasible method in predicting the pork storage time.Practical applicationsFreshness is an important quality characteristic to pork's quality for food industry. Pork storage time is associated with its freshness. From the food safety point of view, it is significant and meaningful to predict the pork storage time. In our research, the proposed Adaboost‐ULDA coupled with Fourier transform near‐infrared reflectance spectroscopy can successfully predict the pork storage time. The results demonstrate the application prospect of this method to the quality control of pork products.