Research Of Nondestructive Detection For Hawthorn Based On Hyperspectral Imaging

As food and traditional Chinese medicine,people favour hawthorn because of its nutritional and medicinal values. This paper researched the fast and nondestructive detecting method for varieties、 qualities、and damage of the hawthorn based on hyperspectral imaging and select three species of hawthorn produced from shuiquan town, zaozhuang city, Shandong province as research objects. It provided theory evidence for the online non-destructive classifying technology.Conclusions of the research are as follows:(1)The Research Preprocessed hawthorn’s spectral data by 13 kinds of pretreatment method and found SGolay(3,2)’result is best. Built PLS model by Spectral data preprocessed by SGolay(3,2), its Rc2 and Rp2 were higher than other 12 kinds of pretreatment method, the RMSEC and RMSEP are less than other 12 kinds of pretreatment method.(2) in the Hawthorn varieties detection, the PLS, PCR, LS-SVM, SPA-PLS, SPA-PCR, SPA-LS-SVM, PCA-PLS, PCA-PCR and PCA-LS-SVM modeling method have carried out the Hawthorn varieties and predicted the prediction set’s data. Comparing the nine models’results, the research found PCA-LS-SVM model built by top 10 principal components works best and its Discriminant accuracy is 98.67%.(3) In hawthorn quality detection, the research built the PLS, LS-SVM, SPA-PLS, SPA-LS-SVM, four comprehensive models about soluble solids, acid, vitamin C, hardness and predicted the prediction set’s data. Comparing the four models’ results, the research found SPA-LS-SVM model built by 22 characteristic wavelengths works best and its Rp2 are 0.8787、0.8316、-0.8946 以及 0.7313.(4) In hawthorn damage detection, the research built models by the principal component, the texture features and the fusion parameters of spectral and image. Among them, discriminant accuracy of PLS,PCR,LS-SVM model built by principal components were 100%,98%,100% and discriminant accuracys of LS-SVM models built by texture features and the fusion parameters were 100%.It proved the LS-SVM models built by the principal component, the texture features and the fusion parameters of spectral and image can distinguish hawthorn damage.