Abstract: In recent years, incidents of sliced mutton adulteration such as adulteration with non-mutton ingredients and the use of restructured and processed meat products have occurred frequently in the consumer market, harming consumer interests and disrupting market order. Existing detection methods suffer from shortcomings such as lengthy detection period and complex sample processing. To address these issues, this study proposed an image recognition approach integrating smartphone shooting systems with chemometrics. The optimal models were developed for high-precision identification of frozen whole-cut, processed, and reconstituted mutton slices. This study extracted 23 features from each of the three kinds of sliced mutton, including mean values and standard deviations of each channel in different color spaces, along with homogeneity, correlation, contrast, energy, and entropy from the grayscale co-occurrence matrix. After dimensionality reduction by principal component analysis (PCA), classification models were established using K-nearest neighbors (KNN), linear discriminant analysis (LDA), random forest (RF), and support vector machine (SVM). Results indicated that the RF model, with a classification accuracy of 91.67%, demonstrated superior overall performance compared with the other three models. SVM and KNN also demonstrated relatively robust classification performance, whereas the LDA model struggled to effectively handle the complex category boundaries of mutton samples, resulting in weaker classification outcomes. The findings of this study confirm the feasibility of using smartphone images combined with machine learning and chemometric methods for identifying adulterated mutton slices.

Key words: sliced mutton; adulteration detection; machine vision; chemometrics; mobile phone images