Identification of beef adulteration based on near-infrared spectroscopy and an ensemble of radical basis function network

Abstract

Driven by huge market demand and economic benefits, the counterfeiting of beef is becoming increasingly rampant. Developing sensitive, accurate, and rapid detection techniques of beef identification and adulteration is of great significance. The present work aims at exploring the feasibility of combining near-infrared (NIR) spectroscopy with pattern recognition for identifying the beef adulterated with pork. In the frame of ensemble learning, two radical basis function (RBF) networks-based ensemble algorithm, abbreviated as “ERBF” and “SERBF”, were designed. Classic partial least squares (PLS), single RBF network were also used for comparison. A total of 212 samples including pure beef and adulterated samples were prepared. Principal component analysis (PCA) was applied for exploratory analysis. The recognition result can be obtained based on the sample spectrum and the corresponding model. On the test set, the SERBF model was shown to provide the best performance with 91.9 %, 95.7 % and 94.3 % of accuracy, sensitivity, and specificity, respectively. This result revealed that the SERBF combined with NIR spectroscopy may be an alternative to traditional methods for quality control of beef.