文献类型： 外文期刊

作者： Yu, Min ¹ ; Yan, Jia ¹ ; Chu, Jiawei ¹ ; Qi, Hang ¹ ; Xu, Peng ¹ ; Liu, Shengquan ¹ ; Zhou, Liang ¹ ; Gao, Junlan ² ;

作者机构： 1.Anhui Agr Univ, Sch Mat & Chem, Key Lab Natl Forestry & Grassland Adm Wood Qual Im, Hefei 230036, Peoples R China

2.Anhui Acad Agr Sci, Inst Agr Engn, Hefei 230031, Peoples R China

关键词： Terahertz time-domain spectroscopy; Moisture content; Competitive adaptive reweighted sampling; Shapley additive explanation; XGBoost

期刊名称：INDUSTRIAL CROPS AND PRODUCTS （ 影响因子：6.2； 五年影响因子：6.2 ）

ISSN： 0926-6690

年卷期： 2025 年 227 卷

页码：

收录情况： SCI

摘要： Terahertz waves, being highly sensitive to moisture, thus have significant potential in wood moisture content detection. This study utilized terahertz time-domain spectroscopy (THz-TDS) to acquire spectral signals from poplar wood samples with varying moisture contents and extract their absorption coefficients. Classical machine learning algorithms (PLSR, DT, and RF), regularization algorithms (LR, RR, and ENR), and gradient boosting decision trees algorithms (CatBoost, LightGBM, and XGBoost) were then applied to develop predictive models for wood moisture content. Feature selection of the absorption coefficients was performed using the Competitive Adaptive Reweighted Sampling (CARS) method, while grid search and cross-validation were employed to optimize model hyperparameters. The impact of feature selection and hyperparameter optimization on prediction accuracy was assessed, and the Shapley Additive exPlanation (SHAP) method was applied to interpret the optimal model. Results indicated a positive correlation between wood moisture content and the THz absorption coefficient. The gradient boosting decision tree algorithms demonstrated superior predictive accuracy over classical machine learning and regularization algorithms. Feature selection and hyperparameter optimization significantly improved the model's predictive performance. Among these, the XGBoost algorithm provided the best model for predicting wood moisture content, achieving a coefficient of determination (R2) greater than 0.96 in the test set. SHAP analysis provided valuable insights into the contribution of specific terahertz frequencies and identified the 0.286 THz frequency as crucial for predicting wood moisture content. This study demonstrates that terahertz time-domain spectroscopy, combined with machine learning algorithms, offers a fast and accurate method for detecting wood moisture content.