Research on a method for analysing the degree of wood modification based on THz-TDS

The performance enhancement of thermosetting resin-modified wood critically depends on modification degree, making its accurate assessment essential for engineering applications. Terahertz time-domain spectroscopy (THz-TDS) offers non-destructive evaluation advantages through material fingerprinting, but conventional optical-parameter methods face limitations due to computational inefficiency and sensitivity to wood texture, compromising detection stability. To address this, we propose a hyperparameter-adaptive dimensionality reduction and unsupervised analysis framework integrating principal component analysis (PCA) and Barnes-Hut t-SNE for efficient compression of high-dimensional terahertz features. Applied to 10,035 samples with 4,500-dimensional data, dimensionality reduction averaged 6.62 s while maintaining feature separability. Bayesian optimization tuned hyperparameters using the Hopkins statistic, followed by DBSCAN partitioning of textured, non-textured, and boundary regions. Imaging validation (0.2 mm resolution) confirmed the framework’s effectiveness in characterizing subtle internal modification variations. Critically, at 1.42 THz—where absorption differences between textured/non-textured regions minimized to 0.9%—texture influence was weakest. This enables imaging results to dominantly reflect modification information for reliable differentiation, demonstrating the method’s robustness for non-destructive wood modification assessment.