Study On Quality Detection Of Wolfberry Based On Hyperspectral Image Technology

Wolfberry is rich in nutrients.But,its quality is uneven in the market,which damages to the economic interests of farmers and consumers seriously.The reason is that there are limited methods for testing the quality of wolfberry at present.Therefore,it is of great practical significance to study the rapid,nondestructive and accurate method for testing the quality of wolfberry.Hyperspectral image technology combines the advantages of image information and spectral information,and recording the full spectrum of each pixel in the image.Polysaccharides,total sugars and flavonoids in wolfberry are the main active chemical constituents,and they are also the important indexes to evaluate the quality of wolfberry.Therefore,this paper take the use of hyperspectral imaging technology to predict polysaccharide,total sugar and flavonoid content,so as to realize the rapid and nondestructive detection for the internal quality of wolfberry.The main work of this paper is as follows:(1)Research on spectral information1)Compare three common spectral pretreatment methods,multiple scatter correction(MSC),Savitzky-Golay(S-G smoothing),and standard normal variate(SNV),and MSC was selected to process the original spectrum,in order to eliminate the influence of scattering.2)The average spectral reflectance of the effective bands,the visible bands,the near infrared bands and the whole bands were selected as the characteristic parameters,and the results of the four kinds of bands were compared.Results showed that the effective bands can obtain the sensitive spectral bands for predicting the content of polysaccharide,total sugar and flavonoid.3)Compared the four modeling methods,it can be concluded that the prediction effect of nonlinear modeling method shows better.(2)Research on image information1)The original hyperspectral images were processed by principal component analysis firstly.Then the first three principal component images was selected and five characteristic wavelengths were identified according to the weight coefficient distribution curve of full bands in the first three principal components.2)The texture features from principal component images,the color features,texture features and spectral features from the characteristic wavelengths images were extracted simultaneously.Partial least squares regression was used to establish the prediction model of polysaccharide,total sugar and flavonoid content based on different characteristic parameters.The results showed that these models were not ideal for predicting them.(3)Optimization of spectral information and image informationAs to different indicators,througth screening and optimizating characteristic variables according to the correlation coefficient threshold can not only decrease the redundant information and reduce the complexity of computation,but also improve the prediction performance.The discriminant accuracy of polysaccharide was 94.44%,coefficient of determination was 0.92,the RMSE is 0.02.The discriminant accuracy of total sugar is 95.56%,the coefficient of determination is 0.91,the RMSE was 0.32.The discriminant accuracy of flavonoids is 92.22%,the coefficient of determination is 0.90,the RMSE is 78.56.