Research About The Nondestructive Measurement Testing Modeling And Optimization Of Potato Based On Hyperspectral Imaging Technology

The hyperspectral imaging technology is the preferred technology to detect the comprehensive quality of fruits and vegetables with characteristics of continuous multi-band, high spectral resolution and so on. Potato has the priority to the development because it is one of the strategic leading industries of Ningxia. Potato’s quality detection is an important link of potato processing. Based on Hyperspectral Imaging Technology (HIT), Back Propagation-Artificial Neural Network (BP-ANN), Radical Basis Function-Artificial Neural Network (RBF-ANN), Support Vector Machine-Artificial Neural Network(SVM-ANN), Bayes Classifier, Partial Least Squares(PLS), This study maked potato external quality as research subjects to set up modeling of potato external quality and potato external quality endmember, including bug-eyes, dry rot, qualified, abnormal, mechanical damage and green skin. The main research contents and conclusions of this paper are as follows:(1)The interest region of different kinds of sample spectrum are withdrew and analyzed, which provide researchs with a sound theoretical basis for prediction models of potato external quality and potato external quality endmember based on Hyperspectral Imaging Technology.(2)Principal component analysis (PCA) was set up to determine five wavelengths (580nm,650nm,676nm,690nm,960nm) as feature bands. At the same time using principal component analysis (PCA), independent principal component analysis (ICA) and SD methord with original hyperspectral image data of potato, which showed that it can be established and classified with PC2images (PCA for feature bands spectra images, called dimension reduction methord Ⅰ), IC2images (ICA for feature bands spectra images, called dimension reduction methord Ⅱ) and different band ratio images(SD for feature bands spectra images, called dimension reduction methordⅢ).(3)Compared De-noising effect with Recursive of Least Squares (RLS) and Wavelet, Wavelet and Recursive of Least Squares (WT-RLS) is obvious, and which has good reconsitution effect by4scale Haar wavelet transform.(4)BP-ANN, RBF-ANN, SVM-ANN, PLS, Bayes Classifier are applied to set up prediction models of potato external potato external defect with the pretreatment of spectra images, the result turned out the model can create the better prediction mode by dimension reduction methord Ⅲ+WT-RLS method.(5)First Derivative(FD), Second Derivative(SD), Savitsky-Golay(SG), Multiplicative Scattering Correction(MSC), Standard Normal Variate Transformation(SNV), wavelet compression are used for band(500nm～960nm) of the pretreatment of spectra data. And then BP-ANN, RBF-ANN, SVM-ANN, PLS, Bayes Classifier are applied to set up prediction models of the interest region of potato external, it showed that the model can create the best prediction mode by BP-ANN+SG-MSC method, the identification rate reach to86.73%, correlation coefficient(R) is equal to0.95, RMSEP is0.40.(6)Establishing bayes classifier methord can gain the reliable and effective identification model of potato external quality endmember compared with SVM-ANN, the identification rate can reach to86.11%.