Objective: If the dried apple is not dried enough in the production process, it is easy to cause the growth of mold, leading to serious food safety problems. Therefore, it is of great economic and social significance to find a method to detect the moisture content of dried apple efficiently and non-destructively.
Methods：In this paper, fresh apple samples were made, dried in a drying oven, and 8 gradients of different moisture content were made. The difference of terahertz spectra of dried apple with different moisture content was investigated by terahertz absorption spectrum.
Result：In the results of preprocessing, for Partial Least Squares, the optimal model is Normalization- Partial Least Squares, which has an RMSEP of 2.0289 and an RP of 0.8985. For Least Squares Support Vector Machine, the optimal model is 1st Derivative-Least Squares Support Vector Machine, which has an RMSEP of 1.1757 and an RP of 0.9685. After the addition of the feature extraction, it was found that the optimal model is 1st Derivative-Uninformative Variable Elimination-Least Squares Support Vector Machine, which has an RMSEP of 1.0483 and an RP of 0.9761. Compared to the Least Squares Support Vector Machine model of raw data, the RMSEP reduced by 0.3968 and the RP improved by 2.57%.
Conclusion：In this paper, the feasibility of using terahertz spectroscopy to predict the moisture content of dried apples was verified, and a moisture content prediction model with high accuracy was established.