Abstract
To rapidly and accurately detect the quality of blueberry, hyperspectral imaging (HSI) technique was used to simultaneously detect the soluble solids content (SSC) and firmness (FI) of blueberry. In total, 204 blueberry samples, including 164 samples in Calibration set and 40 samples in prediction set, were investigated in this study. Multi-stage successive projections algorithm (MS-SPA) and SPA1/SPA2 were proposed to select a few feature wavelengths from the spectral region of 450–950 nm. Prediction models were developed based on partial least squares regression (PLSR), support vector regression (SVR) and back propagation neural network (BPNN) model. The results showed that prediction model based on MS-SPA performed better in prediction results. Furthermore, the prediction based on BPNN model was better than that based on PLSR and SVR models, which used full spectrum (FS), SPA1/SPA2, MS-SPA, respectively, to select feature wavelengths. This research suggested that MS-SPA-BPNN model, which obtained the best prediction results of SSC (RP = 0.894, RMSEP = 0.220), and FI (RP = 0.843, RMSE = 0.225), was a reliable tool to detect SSC and FI simultaneously. The visualization of distribution map of parameters was an intuitive and convenient measurement for quality detection of blueberry. The method could provide a theoretical basis for developing an online detecting and grading system of blueberry quality based on multispectral imaging technique.
