Research About Ningxia Tan Lamb Internal Quality Detection Method Based On Hyperspectral Technology

In this study, Tan lamb in Ningxia were taken as the experimental subjects, using the near-infrared hyperspectral imaging system (900-1700nm) acquire76lamb samples, proposed BP neural network fusion algorithm combined with chemometric methods, established the quantitative prediction models for the content of lamb fat and protein. Then, the results of the models were evaluated and analyzed. The purpose was to find a new method for detecting the internal quality of meat quickly and nondestructively as well as to provide a theoretical for online non-destructive testing of the products. The main research achievements of the paper are as follows:(1)Detected the sample of fat and protein content, and achieved the raw spectral data. Used a variety of methods for spectral pretreatment, and finally get the best pretreatment method of the model for lamb fat content is wavelet transform method that the wavelet function is db4, decomposition scale is3, and the best pretreatment method of the model for the protein content is SNV+S-G.(2) It reduced the dimension of the hyperspectral image data, chooses the fat content of the four characteristic wavelengths (960nm,1132nm,1300nm,1622nm),and the protein content of the four characteristic wavelengths (990nm,1162nm,1323nm,1633nm) used to create a follow-prediction model.(3)Used PLS, RBF-ANN and LS-SVM build prediction models after the fat and protein content of samples removed outliers, the results show that the RBF-ANN model is the highest accuracy, at the full wavelength RBF-ANN model’s Rfat=0.5987, RMSEPfat=2.4994, Rpro=0.6669, RMSEPpro=2.8173; At the optimal wavelength RBF-ANN model’s R&t=0.8970,RMSEP&t=0.6039,Rpro=0.9127,RMSEPpro=0.7317. Results were better than the other two full-band prediction method.While the PLS model prediction accuracy is better than the LS-SVM model; LS-SVM model precision is the lowest, the correlation coefficient R is more minimum, and the RMSEP relatively max.(4) Established BP-ANN prediction model for the optimized data, and the correlation coefficient (R) and predict the root mean square error (RMSEP) of fat and protein prediction model were Rfat=0.8633, Rpro=0.8963and RMSEPfat=0.8015, RMSEPpro=0.8693; To improve the predictive effect, this paper presents genetic algorithm (GA) and particle swarm optimization (PSO) to optimize BP-ANN prediction model, and the results show that PSO+BP-ANN predict is the best, Rfat=0.9204, Rpro=0.9306, RMSEPfat=0.6451, RMSEPpro=0.5819. BP neural network fusion algorithm can better predict the content of lamb fat and protein, and provide the basis for rapid detection of hyperspectral technology lamb quality.