Potential of Hyperspectral Imaging for Rapid Prediction of Anthocyanin Content of Purple-Fleshed Sweet Potato Slices During Drying Process

This study aimed to investigate the potential of hyperspectral imaging technique in tandem with chemometrics analysis for rapid and nondestructive determination of anthocyanin content within purple-fleshed sweet potato (PFSP) during drying process. Hyperspectral images of PFSP in the spectral range of 371–1023 nm were obtained during contact ultrasound-assisted hot air drying (CUHAD) process, and the reference anthocyanin contents of PFSP were measured by a traditional method. Partial least square regression (PLSR) and least-square support vector machine (LS-SVM) were applied to establish the calibration models based on raw extracted spectrum and spectrum preprocessed by four different methods. In order to simplify the calibration model, three algorithms including PLSR, LS-SVM, and multiple linear regression (MLR) were used to build models based on ten optimal wavelengths selected by regression coefficients (RC) method. The results showed that the RC-MLR yielded best results with the coefficient of determination for calibration (RC2$$ {R}_C^2 $$) of 0.868 and coefficient of determination for prediction (RP2$$ {R}_P^2 $$) of 0.866. Finally, distribution maps were developed based on an image processing algorithm to visualize anthocyanin content of PFSP at different drying periods which cannot be achieved by conventional methods. The overall results demonstrated that hyperspectral imaging technique is a useful tool for rapid and nondestructive determination of the anthocyanin content during drying process.