Non-Destructive Prediction of Moisture Content and Freezable Water Content of Purple-Fleshed Sweet Potato Slices during Drying Process Using Hyperspectral Imaging Technique

The aim of this study was to investigate the feasibility of hyperspectral imaging in measuring moisture content and freezable water content during drying process. Hyperspectral images were acquired for purple-fleshed sweet potato (PFSP) slices during contact ultrasound assisted hot drying (CUHAD) process, and the corresponding mean reflectance spectra from regions of interest in visible and near infrared (371-1023 nm) regions were extracted. Moving average, Savitzky-Golay smoothing filter (S_G filter) and multiplicative scatter correction (MSC) were investigated to preprocess the raw spectra and partial least square regression (PLSR) calibration model was established to analyze the relationship between the extracted spectral data and measured quality attributes. Comparing the performance of model based on different preprocessing methods, the PLSR model with MSC pre-treatment presented better results with coefficients of determination for prediction () of 0.9837 and 0.9323 for moisture content and freezable water content, respectively. Instead of selecting full range spectra data, optimal wavelengths were identified based on the regression coefficients (RC) method. Then two linear calibration algorithms named PLSR and multiple linear regression (MLR), and a non-linear calibration algorithm named backpropagation (BP) neural network were used to establish models to predict quality attributes of samples simultaneously. The results showed that the RC-MLR with of 0.9359 and 0.8592 was considered as the best for determining moisture content and freezable water content of PFSP slices. Therefore, the study demonstrates the potential of using hyperspectral imaging in tandem with chemometrics analysis as an objective, fast and non-destructive method for predicting the moisture content and freezable water content at different dehydrated times.