Abstract: In this experiment, hyperspectral images in different wavelength ranges were acquired for carrot slice samples during hot air drying. Subsequently, using multivariate statistical analysis combined with chemometrics, a predictive model for the non-destructive determination of moisture content (MC) and carotenoid content (CC) in samples was developed separately based on partial least squares (PLS) and support vector machine (SVM) algorithm. The results showed that the SVM models developed using multiplicative scatter correction (MSC) in the 400–1 000 nm had the best prediction performance for both MC and CC with coefficient of determination for prediction (R2P) of 0.984 and 0.911, and root mean square error for prediction (RMSEP) of 0.380 g/g and 34.836 mg/100 g, respectively. The optimal models with the feature wavelengths selected by successive projections algorithm showed R2P of 0.962 and 0.898 and RMSEP of 0.612 g/g and 37.544 mg/100 g for MC and CC, respectively. The residual predictive deviation (RPD) in the new models was over 3, indicating good accuracy and stability. Moreover, the spatial distribution of moisture and carotenoid during the drying process were generated and visualized as pseudo-color images. The results indicated that the hyperspectral imaging could be used to effectively predict the MC and CC in carrot slices, demonstrating the potential of hyperspectral imaging as an analytical tool in quality control of carrot slices during drying.

Key words: carrot slices, drying, moisture content, carotenoid content, hyperspectral imaging, visualization