Prediction model for moisture content in cantaloupe slices using LF-NMR and different drying methods

Abstract: Intelligent drying is one of the most promising directions to preserve a wide variety of food and agricultural products in modern drying technology. It is necessary to real-time monitor the physicochemical properties of materials in the drying process, thereby regulating the drying parameters. Furthermore, the water state and the moisture content of materials are the key information to control the drying process. In this study, the hot air drying (HA) and infrared drying (IR) were used at the same level of temperature (50, 60, and 70℃) to dehydrate the cantaloupe slices. A low-field nuclear magnetic resonance (LF-NMR) was utilized to analyze the moisture migration and its variation in the products during two drying processes. A robust prediction model of moisture content was established using chemometric methods, where the NMR parameters of samples were obtained by LF-NMR during HA and IR. The results showed that the drying temperature significantly affected the drying characteristics of cantaloupe slices, either HA or IR. The high temperature was beneficial to shorten the drying time, due mainly to the high efficiency of mass transfer during drying. A higher drying efficiency was achieved in the IR, compared with HA. The reason was that the infrared radiation in the IR can penetrate the material to realize internal heating. The IR shortened the drying time up to 20.0%-37.5%, at the same temperature level. In LF-NMR analysis, the bound water, immobile water and free water were detected in fresh cantaloupe. The curves of transverse relaxation time (T2) moved to the left of the coordinate during HA and IR process, indicating the degree of freedom of moisture was reduced in the sample. There were also some changes in the NMR parameters obtained from the T2 curves. Specifically, the peak of transverse relaxation time for the free water T23p decreased steadily, whereas, that for the immobile water T22p showed various change trends, due to the differences in drying method and drying temperature. Unlike the gradual decrease in T21p (the peak of transverse relaxation time for the bound water) during the HA process, T21p raised briefly at the initial stage of IR and then decreased. This phenomenon can be due to the vibration and rotation of organic molecules in the material, when absorbed the infrared energy, which caused the state change of water that was combined with the organic molecules. The peak area of free water A23 gradually decreased, while that of immobile water A22 and bound water A21 fluctuated in the HA and IR. The A22 and A21 showed an upward trend before the disappearance, indicating that it was related to the conversion among water with different states. There was only a peak of bound water in the sample at the end of drying. The collected moisture content and NMR parameters during drying were used to establish the univariate model, multiple linear regression model (MLR), partial least squares regression model (PLSR), and multiple nonlinear regression model (Support vector machine, SVM) for the prediction of moisture content in the cantaloupes. The best performance with a determination coefficient of 0.986 (calibration set) was achieved in the PLSR model suitable for collinearity problems, compared with the MLR and SVM models. Furthermore, the determination coefficient for predicting (R2P) of the PLSR model independent of HA or IR dataset was higher than 0.99, indicating that the PLSR combined with LF-NMR can realize the rapid determination of moisture content for cantaloupe slices, while it was not affected by the variances in water status caused by different drying. This finding can provide an insightful basis for the prediction model of moisture content in the fruits and vegetables using LF-NMR and multi-processing.