Online analysis of watercore apples by considering different speeds and orientations based on Vis/NIR full-transmittance spectroscopy

Abstract

Watercore is a common internal disorder of ‘Fuji’ apples, which affects the quality and price of fruit. Moreover, the flesh of watercore apples is prone to browning during storage, resulting in a loss of commercial value. However, the online detection of watercore apples is very difficult in actual production due to the interference of many factors. In this study, visible and near infrared (Vis/NIR) full-transmittance spectroscopy (680-1000 nm) was used to online analyze watercore apples. Three different detection orientations (O1, O2 and O3) and speeds (S1, S2 and S3) were compared in detail. In order to determine the optimal detection speed and orientation, the partial least square discriminant analysis (PLS-DA) and least squares-support vector machine (LS-SVM) models were established based on the preprocessing spectra of Savitzky-Golay smoothing and standard normal variate (SGS-SNV). The results showed that S2 speed (0.5 m/s) and O3 orientation (apple stem-calyx axis horizontal and parallel to the moving direction of conveyor belt) were the most suitable for detection of watercore apples. A combination algorithm (MC-UVE-SPA) of Monte Carlo-uninformative variable elimination (MC-UVE) and successive projections algorithm (SPA) was used to select effective wavelengths for classification of watercore apples and quantitative prediction of watercore degree. MC-UVE-SPA-PLS-DA and MC-UVE-SPA-LS-SVM models established based on effective wavelengths obtained the same classification performance with the success rates of 100 % and 96.87 % for healthy and watercore apples, respectively. In order to predict the watercore degree of apples, the partial least squares (PLS), multiple linear regression (MLR) and LS-SVM models were established based on spectra obtained S2 speed and O3 orientation. MC-UVE-SPA-LS-SVM model coupled with eight effective wavelengths obtained the optimal prediction accuracy of watercore degree with 0.93 of RP and 2.12 % of RMSEP. The overall results indicated that online detection of watercore apples based on Vis/NIR full-transmittance spectroscopy was feasible, and the classification accuracy was related to the detection speed and the sample orientation. In addition, the results also showed that the LS-SVM model has good performance for classification of watercore apples and quantitative evaluation of watercore degree.