Apple Variety Identification Based on Dielectric Spectra and Chemometric Methods

Abstract

The dielectric properties of 160 apples of three varieties were obtained from 10 to 1,800 MHz. Based on the Kennard-Stone algorithm, 106 apples were selected for calibration set and the remaining 54 apples were used for validation set. Principal component analysis (PCA) and successive projections algorithm (SPA) were used to extract characteristic variables from original full dielectric spectra (FS). The learning vector quantization (LVQ) network, support vector machine (SVM), and extreme learning machine (ELM) modeling algorithms were applied to build models to identify the varieties of apples. Results showed that the first three principal components, and two dielectric constants and ten loss factors were selected as characteristic variables by PCA and SPA, respectively. SPA-ELM and PCA-ELM, whose total average accuracy reached 99.5 and 99.0 %, respectively, had good potential in identifying apple varieties. The study indicates that the dielectric spectra with chemometrics are promising for identifying apple varieties nondestructively and accurately.