Non-destructive Detection Of Chilled Mutton Quality Based On Byperspectral Imaging Technology

In this study, the mutton quality parameters of Tan Han crossbreed sheep in terms of tenderness, chromaticity, moisture content, storage time and the discriminant of chilled and frozen meat were evaluated by hyperspectral imaging technology (400-1000 nm) combined with chemometrics method, which can provide the theoretical and technical support for on-line detection of animal products. The main reseaRch contents and results are as follows:(1) Compared to the original spectrum model, PLSR models established by 400-1000 nm spectrum with Savitzky-Golay smoothing showed better performance to predict the tenderness in chilled mutton. The correlation coefficient of calibration and validation about tenderness prediction models based on 400-1000 nm spectrum with Savitzky-Golay smoothing were 0.833 and 0.865. The correlation coefficient of calibration and validation about color L* value (luminance), color a* value (redness) and color b* value (yellowness) prediction models based on 400-1000 nm spectrum with Savitzky-Golay smoothing, SNV and MSC were 0.834,0.834,0.904 and 0.736,0.796,0.898. Compared to the original spectrum model, PLSR models established by 400-1000 nm spectrum with pretreatment approaches presented excellent ability to predict the color parameter in chilled mutton.(2) Compared to PCR models of the original spectrum, PLSR models based on 400-1000 nm spectrum have an excellent ability to predict the moisture content in chilled mutton, with which correlation coefficient of calibration and validation were 0.867 and 0.749, root mean square error was 0.707. Compared to the original spectrum model and pretreated spectrum models, PLSR models established by 400-1000 nni spectrum with Savitzky-Golay smoothing produced the best effect to predict the moisture content in chilled mutton, with which correlation coefficient of calibration and validation were 0.888 and 0.784, root mean square error was 0.696.(3) The LDA model was established for the whole band spectrum of 400-1000 nm. When the number of principal components was PC-15, the model was best to determine the cold storage time of mutton, and the total discriminant accuracy was 98.13%. For a model in the case of constant principal components, the effect of model based on the original spectrum with 6 points Peak normalization was better. To the LDA model established based on the 400-1000 nm spectrum with 6 points Peak normalization, the total discriminant accuracy rate of the model was 91.25%, which effect of model was better than the original spectrum and other pretreated spectrum models. Compared to the effect of the original spectrum model with proper principal components of PC-15 and the original spectrum model with 6 points Peak normalization, the results showed that the effect of the principal component on the LDA discriminant model is bigger than that of the spectral pretreatment methods.(4) For discrimination of chilled and frozen meat, the PLSR-DA model established using theoriginal spectra with MSC was found to have better effect for distinguishing chilled meat and frozen meat than the original spectrum and other pretreatment spectrum models, with which correlation coefficient of calibration and validation were 0.989 and 0.991, root mean square error of calibration and validation were 0.149 and 0.187. Nineteen characteristic wavelengths (PC1+PC2+PC3) which were optimized by principal component weighting coefficient can be used to replace the whole band spectrum to discriminant chilled and frozen meat by PLSR-DA models, with which correlation coefficient of calibration and validation were 0.982 and 0.997, root mean square error of calibration and validation were 0.191 and 0.230.Seven characteristic wavelengths were selected using Successive Projections Algorithm (SPA) method, the recognition accuracy of models developed by support vector machine (SVM) method with whole bands and characteristic bands was 100%. Which indicated that characteristic wavelengths selected by SPA method had higher effectiveness. Therefore, the characteristic wavelengths can completely replace the whole band to discriminant chilled and frozen meat by SVM model. Compared to the PLSR-DA model, SVM discriminant model had higher identification accuracy to classify chilled and frozen meat, and was more suitable for rapid discrimination of chilled and frozen meat.