Internal Quality Classification of Apples Based on Near Infrared Spectroscopy and Evidence Theory

Abstract

AbstractApple classification is of great significance to enhance market competitiveness of Chinese apple industry in the world. In this paper, the collected apple near infrared spectroscopy is taken as the data sample. Firstly, data processing was carried out on collected samples: Principal Component Analysis – Mahalanobis Distance method was used to eliminate abnormal samples, convolution smoothing filtering was used to remove the noise in the original spectrum, Multiple Scattering Correction and Standard Normal Variate were used to calibrate the baseline of apples spectra, Genetic Algorithm was used to eliminate the invalid wavelength information, and prediction models of Extreme Learning Machine and Partial Least Squares were established. In order to solve the problem of classification accuracy decline caused by hard segmentation, uncertainty was introduced, that is, based on evidence theory apple classification fusion algorithm. By assigning different discount factors to quality functions of different prediction models, a new basic probability function is generated, and then the data fusion of the new basic probability function is carried out by using evidence combination rules and it improves the accuracy of apple classification.KeywordsApple classificationExtreme learning machinePartial least squaresEvidence theory