Nondestructive Measurement of Soluble Solids Content of Kiwifruits Using Near-Infrared Hyperspectral Imaging

Abstract

This study aims to determine the soluble solids content (SSC) of intact kiwifruits of varieties “Xixuan” and “Huayou” by using near-infrared (NIR) hyperspectral imaging and to investigate which model, developed for a single variety or for two varieties together, had better SSC determination performance. The NIR hyperspectral reflectance images of 200 kiwifruits (100 kiwifruits for each variety) were obtained over the wavelength of 865.11–1,711.71 nm. Mean spectra were extracted from the regions of interest in hyperspectral images of each kiwifruit. The samples were divided into calibration set and prediction set based on the joint x–y distance sample set partitioning method. There were 67 samples in calibration set and 33 samples in prediction set when a single variety was used to establish SSC calibration model, and there were 134 samples (67 “Xixuan” and 67 “Huayou”) in calibration set and 66 samples (33 “Xixuan” and 33 “Huayou”) in prediction set when the two varieties were used together. Successive projections algorithm (SPA) was applied to extract the effective wavelengths (EWs) from full spectra (FS). Nine EWs were selected when a single variety (“Xixuan” or “Huayou”) was used in modeling, and 19 EWs were selected when the two varieties were used together. SSC calibration models were developed based on the partial least squares (PLS) regression and least square support vector machine (LSSVM) modeling methods using the full spectra and extracted EWs as inputs, respectively. The results showed that both calibration and prediction performances of LSSVM models were better than those of PLS. The best SSC determination model for “Xixuan,” “Huayou,” and two varieties together were SPA-LSSVM, FS-LSSVM, and FS-LSSVM with the correlation coefficient of prediction set of 0.766, 0.971, and 0.911, and the root-mean-square error of prediction set of 0.968, 0.589, and 1.137, respectively. The study demonstrates the feasibility of using NIR hyperspectral reflectance imaging technique as a noninvasive method for predicting SSC of kiwifruits and indicates that developing a model for a specific variety is helpful to decreasing prediction error and to improving calculation speed.