Integrating hyperspectral and terahertz spectroscopy for enhanced wood species recognition

Wood is a valuable natural resource, and effective utilization requires accurate species identification. This study presents a method for wood species recognition using hyperspectral (400–2500 nm) and terahertz (30–3000 μm) spectroscopy combined with data fusion strategy. Ten species, including five broadleaf (Xylosma racemosum, Populus davidiana, Fraxinus rhynchophylla, Betula platyphylla and Tilia tuan Szyszyl) and five coniferous (Pinus sylvestris, Pinus tabulaeformis, Pinus massoniana, Larix gmelinii and Pinus koraiensis) species, were selected as experimental samples. Spectral data were collected and fused at low and mid-levels. The spectra were preprocessed using Savitzky-Golay smoothing, standard normal variate (SNV), and multiple scattering correction (MSC). Subsequently, support vector machine (SVM) and extreme learning machine (ELM) models were developed for wood species recognition. The results show that data fusion significantly improved the recognition accuracy. For broadleaf species, the recognition rate reached 100%, while for coniferous species, SVM with SG preprocessing and low-level data fusion achieved 96%, a notable improvement over single-spectral models. These findings demonstrate the complementary roles of visible-near-infrared and terahertz spectroscopy for wood species identification.