Study Of Non-destructive Inspection For Frozen Food Based On Hyper-spectral Imaging Technology

Beef is a kind of important meat in the daily life for its delicious and high nutritional value. Spoilage usually occurs to meat under the combined action of the enzyme and microbial. Therefore, quality and safety testing to beef is significant before entering the market. At present, the quality of beef is mainly tested by traditional detection methods including sensory evaluation, physical and chemical analysis. It is not only time-consuming, laborious and destructive, but also requires highly skilled operators. Thus, it is not convenient to the rapid detection of beef products. In recent years, hyper-spectral imaging technology combining spectroscopy, image technology and computer vision technology get rapidly development. It provides a new approach for food quality testing in food field.In this thesis, fresh beef, fast frozen-thawed beef and slow frozen-thawed beef were taken as the study object. The texture characteristics, color, moisture content and thawing loss were taken as the evaluating indicator. This dissertation focused on the application of hyper-spectral imaging technique, combined with stoichiometry, image processing methods and computer technology to establish the mathematical models for quick and non-destructive test of beef quality. The main results were as follows:(1) The samples of beef processed in different freezing process were investigated by experiments. The results showed that although freezing treatment affects the quality of beef, it is still able to maintain the quality of beef. The indices of the fast frozen beef were closer to these of fresh beef. So fast freezing process could maintain the quality of beef better.(2) Based on the optimal wavelength, partial least squares regression(PLSR) and multiple linear regression(MLR) models were established to predict the thawing loss of beef processed by different processes, respectively. The results were obtained by PLSR model with correlation coefficient(RP) of 0.929, root mean square error of prediction(RMSEP) of 1.48 and residual predictive deviation(RPD) of 3.95. The results were obtained by MLR model with correlation coefficient(RP) of 0.72, root mean square error of prediction(RMSEP) of 2.56 and residual predictive deviation(RPD) of 2.51. The results of PLSR model were better than these of MLR model. The accuracy and stability of PLSR model are higher than these of MLR model.(3) The least-square support vector machine(LS-SVM) classification model was established based on the optimal wavelength. The results showed that the model could categorize beef samples with different freezing treatment, and its correct classification rate was 83.3%. The correct classification rate and accuracy of the model were enough to categorize beef samples with different freezing treatments.(4) Based on the image texture feature and spectral information, a back propagation artificial neural network(BP-ANN) model was established to predict moisture content of the beef treated by different freezing treatments. The model realized the visualization of the distribution of moisture content of beef with different processing. The result showed that the correlation coefficient(RP) was 0.9054, and the root mean square error of prediction(RMSEP) was 1.48. The model prediction accuracy and the stability were fine.