Abstract: Adulterated beef mixed with pork was qualitatively and quantitatively studied by using an electronic nose combined with statistical analysis. The feature values were extracted by cluster analysis and mean value method. Principal component analysis (PCA) and linear discriminant analysis (LDA) were used for qualitative analysis. Quantitative models were established by partial least squares (PLS), multivariate linear regression (MLR) and BP neural network (BPNN) to quantitatively predict pork adulteration in beef. The results showed that the characteristic value extracted by cluster analysis could reflect the response signal of the electronic nose more comprehensively, while LDA was more suited for qualitatively detecting adulterated beef. The correct classification rates for the training and verification sets were 98.8% and 97.4%, respectively in the multi-layer perceptron neural network analysis, indicating that the classification results are good. The coefficient of determination (0.999 3 and 0.993 0) and root mean square error (0.90% and 2.50%) of the BPNN model were significantly better than those of the other models. Thus, the BPNN model allowed better prediction of the content of pork in adulterated beef. These results conclusively show that an electronic nose is an feasible approach for the detection of adulterated beef mixed with pork.

Key words: beef, pork adulteration, electronic nose, statistical analysis