Study On Different Varieties Of Mutton Recognition Method Based On Hyperspectral Technology Research

This article takes Tan Han crossbreed sheep, Yanchi Tan Sheep and Small Tail Han Sheep as research objects respectively by using 400-1000nm visible near-infrared and 900-1700nm near-infrared hyperspectral imaging technology to collect the hyperspectral images, and by using LDA, SVM, PLS-DA and HCA and other chemometrics methods to make models to identify types of lambs.Here are the main results of those researches:(1) Do SG、SNV、MSC、OSC、Area normalize. Max normalize、SG derivative、FD、SD、 Baseline、Deresolve to the spectrum, which the original spectrum in the range of 400-1000nm and 900-1700nm. Do LDA to the spectrum after the pretreatment, and choose the best option to preprocess the spectrum by using the LDA ratio of the LDA model and build the model.Researches show that: comparing the models made by original spectrum and other 11 attempts in the ranges of 400-1000nm, BC is the best way to build the model. It has a classification rate of 99.44%, the prediction sets ratio is 96.67%. Comparing the models made by original spectrum and other 11 attempts in the ranges of 400-1000nm, SG is the best way to build the model. It has a classification rate of 98.89%, the prediction sets ratio is 100%.(2) Do SVM varieties of discriminant analysis on those three types of lamb in those two ranges of spectrum, and the results show that the classification rate of those lambs is 95% in the range of 400-1000nm, and the classification rate of those lambs is 96.33% in the range of 900-1700nm.(3) Do HCA to the two ranges of original spectrum, and the results show that the HCA has significant effect on full-wave band, and it has a more significant effect on 400-1000nm spectrum than 900-1700nm.(4) Do PLS-DA to the 900-1700nm original spectrum, From all parameters of the model, the Rc is 0.9594, Rcv is 0.9339, Rp is 0.9559, and they have high correlation. The difference between Rc and Rcv is only 0.0255, the RMSEC and RMSEP are 0.23 and both low, which implies that the model has high stability. Besides that, the RMSEC and RMSEP of the calibration model set and the prediction model set are both no more than 0.5, which means that based on all the data,0.5 units above or below the data for mutton categories as discrimination on the basis of the types of mutton can be effective.(5) Do data dimension reduction by PCA to the spectral information of the three types of lamb samples in the range of 400-1000nm and 900-1700nm. And choose six characteristic wavelength from 400-1000nm by weighting coefficient: 990nm,1028nm,1103nm,1312nm,1440nm,1642nm, choose seven characteristic wavelength from 900-1700nm: 410nm,545nm,574nm,631nm,732nm,799nm.(6) Compare the models of original spectrum and characteristic wavelength of the spectral information of those three mutton samples built by Hyper-spectral image, and by using LDA, SVM, and HCA to do the variety of qualitative discriminant analysis. The results show that the LDA model of full-wave band excels the characteristic wavelength’s, and the characteristic wavelength LDA model in the range of 900-1700nm excels those in the range of 400-1000nm. The SVM model made with full-wave band excels the characteristic wavelength’s, and SVM model made with full-wave band in the range of 900-1700nm excels those made in the range of 400-1000nm, the Nu-SVC type of SVM model excels the C-SVC type of SVM models as well, and the SVM model made by linear kernel is the best. By comparing models made by SVM, LDA and HCA, the identification effect of the LDA model excels the SVM and HCA models.