Wood Recogition Based on Terahertz Spectrum and Hyperspectral Technology

Abstract

Terahertz time-domain spectroscopy (THz-TDS) and hyperspectral technology were used for wood recognition. As wood is a valuable national resource, it is essential to utilize it efficiently and reasonably by classifying the species of wood. To accomplish this, ten distinct species of wood, including five types of softwood (Pinus sylvestris, Pinus tabulaeformis, Pinus massoniana, Larix gmelinii, and Pinus koraiensis) and five types of hardwood (Xylosma racemosum, Populus davidiana, Fraxinus rhynchophylla, Betula platyphylla, and Tilia tuan Szyszyl), were selected as experimental samples. Four hundred groups of data for terahertz absorption coefficient spectra and four hundred groups of hyperspectral data were acquired using THz-TDS and hyperspectral technology, respectively, and then examined for their spectral features. Three spectral preprocessing techniques, including the Savitsky–Golay smoothing algorithm, standard normal variable transformation, and multivariate scattering correction, were employed to preprocess the spectrum. Support vector machine recognition models were then created to compare and analyze the effects of recognition. The results demonstrated that both THz-TDS and hyperspectral approaches could successfully identify five different species of hardwood from various families and genera, with the highest accuracy rates of 92 and 94%, respectively. THz-TDS achieved a 92% recognition rate for five different species of softwood from the same family, indicating good recognition effects, while hyperspectral technology did not achieve such results.