Research On Non-destructive Measurement And Fuzzy Recognition Of Apples' Internal Quality Based On Nirs&ls-svm

Non-destructive measurement and recognition of fruits'internal quality is a hot research topic in the area of agricultural and food engineering technology, which will be significant to meet the requirement of food quality and safety, to improve the market value and market competition, and to increase the income of the farmer. At present, conventional analytical methods used for fruits'internal quality are destructive, time- consuming, expensive and low precision, and therefore it is a important work to find non-destructive measurement method for fruits'internal quality based on near infrared spectroscopy and to build the mathematic models and evaluation system of stable performance.This paper chose apples'internal quality as the research object,and selected the sugar content and acidity as the reference index of the apples'internal quality. Areas involving acquisition of apples' near infrared spectroscopy, extraction and analysis of the spectral information, establishment and comparison of the mathematical models, fuzzy recognition of the apples'internal quality were systematically researched. On this basis, the model of non-destructive measurement and fuzzy recognition of apples'internal quality, based on least squares support vector machine, had been established. The major work is summarized as follows:Firstly, the principles and methods of the non-destructive measurement based on near infrared spectroscopy were clarified. Then the measurement platform of apples'near infrared diffuse reflectance spectroscopy was established. The influences of measurement distances and testing parts for the spectral response were explored in order to find the best conditions to get more accurate diffuse reflectance spectra of the apples.Secondly, methods of analysis and extraction for spectral information were studied, including the spectra pretreatments of three ways involving smoothing, derivative and multiplicative scatter correction, outliers elimination based on the model stability and feature extraction depending on principal component analysis and kernel principal component analysis. By means of analysis and extraction for spectral imformation of the apples, the spectral feature dimensions were reduced under the premise of retaining the useful information, which laid the foundation for the future establishment of mathematical models.Thirdly, the linear and nonlinear mathematical models of apples'sugar content and acidity were established based on near infrared spectroscopy. The simulation results showed that the prediction accuracy of the linear model for apple's internal quality was not satisfied, while the non-linear back propagation artificial neural network model improved the prediction accuracy, however, there still existed the problems of unstable of the training process and shortage of generalization ability. The least squares support vector machine algorithm based on structural risk minimization principle was applied to establish a non-linear regression model, to find the optimal model parameters by the cross- validation based on two-grid search and to improve the sparsity and robustness by optimizing training.This model obviously showed the better predictive performance than the back propagation artificial neural network model.On account of the fact that there was no evaluation method available for relevant description of the status of apples'internal quality, the concept of apples'fuzzy membership, which was the Combinations of the data description of membership based on apples' near infrared spectroscopy and the internal quality membership based on the scores of sugar and acidity, was introduced to characterize the apples'internal quality more explicitly and objectively. The constraining programming of fuzzy chance and fuzzy characteristics were researched to establish the fuzzy recognition model of apples'internal quality through the least squares support vector machine algorithm. Apples could be classified according to the classification rule. Through the fuzzy membership rule, the extent of the classes for the apples could be understood clearly. Consequently apples'internal quality was preferably identified.