Effectively Predicting Soluble Solids Content in Apple Based on Hyperspectral Imaging

It is very important to extract effective wavelengths for quantitative analysis of fruit internal quality based on hyperspectral imaging. In the present study, genetic algorithm (GA), successive projections algorithm (SPA) and GA-SPA combining algorithm were used for extracting effective wavelengths from 400 similar to 1000 nm hyperspectral images of Yantai "Fuji" apples, respectively. Based on the effective wavelengths selected by GA, SPA and GA-SPA, different models were built and compared for predicting soluble solids content (SSC) of apple using partial least squares (PLS), least squared support vector machine (LS-SYM) and multiple linear regression (MLR), respectively. A total of 160 samples were prepared for the calibration (n=120) and prediction (n=40) sets. Among all the models, the SPA-MLR achieved the best results, where R-p(2), RMSEP and RPD were 0. 950 1, 0. 308 7 and 4. 476 6 respectively. Results showed that SPA can be effectively used for selecting the effective wavelengths from hyperspectral data. And, SPA-MLR is an optimal modeling method for prediction of apple SSC. Furthermore, less effective wavelengths and simple and easily-interpreted MLR model show that the SPA-MLR model has a great potential for on-line detection of apple SSC and development of a portable instrument.