Study On Nondestructive Detection Of Jujube Quality Based On Hyperspectral Imaging Technology

Hyperspectral imaging technology becomes an important tendency that applied on the fruit quality nondestructive detection. It takes the advantage of image and spectral technology together, has a fast and nondestructive detection on the fruit quality both internal and external. Jujube has become the characteristic production in Ningxia, it attracts much attention.In the paper, it has a nondestructive research on the various parameters based on the internal and external qualities of Jujube by the hyperspectral imaging technology, builds a prediction model in the view of sugar content and the flesh firmness of the long jujubes adopting both the partial least squares method and BP neural network; meanwhile, sets a identification model of the kinds through the supporting vector machine and the BP neural network, and has a comparative analysis on this two models; detects the external defects (subtle bruise and insect damage) of the jujube using the principal component analysis and band ratio algorithm of the characteristic wave band while combining the image processing technology.These detection algorithms offer theoretical basis for developing a real-time、fast and on-line nondestructive detection system.The main research contents are as follows:First, Multiple scattering correction method is used for pretreating the original reflect spectral data, it has a better performance for the prediction and identification model.Second, the model of the sugar content and the flesh firmness of the jujub built by the BP neural network is superior to the partial least square method. The correlation coefficient and the mean square root are0.9205and1.7482of prediction model for sugar content by BP neural network,while the values are0.9042and15.1631respectively of prediction model for flesh firmness by BP neural network.Third, it reduces the dimension of the hyperspectral image data by PCA, and chooses four characteristic wavelengths (1029nm、1229nm、1387nm、1467nm) so as to build a image system based on the filters and apply to the system of on-line nondestructive detection of jujube.Forth, the characteristic wave band of PCA and band ratio algorithm can detect the defects effectively and the whole accuracy of90%.Fifth, it has a better result that builds the identification model of the jujube variety by the supporting vector machine with the radial basis kernel function, the penalty parameter c and the kernel parameter g are obtained by cross-validation,the identification rate can reach to95%.