Characterization of hot air drying behavior and dynamic moisture prediction in goji berries using LF NMR

This study investigated moisture migration in goji berries ( Lycium barbarum L.) during drying using low-field nuclear magnetic resonance (LF NMR) and magnetic resonance imaging (MRI). Drying was conducted at 45, 50, 55, 60, and 65 °C, with relaxation spectra and proton density-weighted images collected to characterize internal moisture distribution and phase transitions. The results showed that increasing temperature significantly shortened drying time and enhanced effective moisture diffusion, thereby accelerating the diffusion and interconversion of free, immobilized, and bound water within goji berries. MRI revealed that water was mainly concentrated in the fruit center, showing an inward-to-outward diffusion pattern. Among six drying kinetic models, the Page model best described moisture loss, with a coefficient of determination ( R 2 ) of 0.9964 and a mean coefficient of variation of 0.2594. LF NMR relaxation parameters and color features ( L *, a *, b *) were used to construct partial least squares regression (PLSR) and convolutional neural networks (CNN) models to predict the dry basis moisture content. The CNN model performed excellently, achieving a prediction set R 2 of 0.9646 and a root mean square error ( RMSE ) of 0.3170, outperforming the PLSR model. Overall, LF NMR combined with deep learning provides accurate prediction, thereby supporting optimization and control of goji berry drying.