提出了一种利用太赫兹时域光谱技术（THz-TDS）进行红木分类识别方法。 红木价格昂贵, 同时由于种类繁多难以识别, 导致红木市场以次充好, 以假乱真的现象层出不穷, 严重扰乱了市场秩序, 给生产者和消费者造成巨大的经济损失, 传统的红木分类识别方法难以兼顾准确性和快速性, 因此需要研究一种新的方法对现有木材分类识别方法进行补充和发展。 相比于传统方法, 太赫兹波对红木具有良好的穿透性及指纹特性, 在红木的分类识别中有较大的应用潜力。 选用5种红木（巴里黄檀、 奥氏黄檀、 大叶紫檀、 小叶紫檀、 交趾黄檀）作为试验样品木材。 利用THz-TDS系统得到木材的太赫兹时域光谱, 通过对五种木材的太赫兹时域光谱进行快速傅里叶变换, 得到木材太赫兹频域光谱, 并对太赫兹时域光谱提取光学参数, 分别得到木材的太赫兹折射率谱和吸收系数谱, 结果表明不同种类的木材在时域光谱上具有时间延迟线与振幅的差异, 在频域光谱上显示衰减趋势及幅值各不相同, 在吸收系数谱中各种类红木吸收峰出现的频段不同, 能够直观地展示出各种类木材之间的区别, 表明THz-TDS进行红木分类识别具有一定的可行性。 利用连续投影算法（SPA）提取吸收系数谱和折射率谱的特征频率, 对吸收系数谱260个频率点筛选出28个特征频率点, 频段占比1077%; 对折射率谱260个频率点筛选出12个特征频率点, 频段占比462%。 分别建立基于吸收系数谱和折射率谱的随机森林分类模型和支持向量机（SVM）红木分类模型, 并对各模型分类结果进行对比。 实验结果表明, THz-TDS具有良好的木材识别效果, 基于木材太赫兹吸收系数谱和折射率谱建立的随机森林分类模型对红木种类有着较好的分类性能, 总体分类准确率分别达到了94%和96%, 能够准确对红木种类进行分类识别。 利用太赫兹时域光谱技术实现了红木的分类识别, 为红木的分类识别提供了一个新的思路和技术方案, 能够作为近红外光谱木材检测方法的补充, 同时为太赫兹技术在木材分类识别领域的应用提供了理论基础。

This paper proposes a method to classify and recognize redwood using Terahertz time-domain spectroscopy (THz-TDS). Redwood is expensive and difficult to identify which leads to a shoddy market. The phenomenon disrupts the market order and causes huge economic losses to producers and consumers. The traditional methods of identifying redwood are difficult to give consideration to both accuracy and rapidity, therefore it is necessary to put forward a new method to supplement the traditional classification methods. Compared with the traditional methods, terahertz wave has good penetrability and fingerprint characteristics for redwood, and has great application potential in classification and identification of redwood. In this paper, five kinds of redwood (Dalbergia bariensis, Dalbergia oliveri, Bois de rose, Pterocarpus santalinus, Dalbergia cochinchinensis) are selected as test samples. The THz-TDS system is used to obtain the terahertz time-domain spectrum of wood; the terahertz frequency domain spectrum is obtained by fast Fourier transform of the terahertz time-domain spectrum of five woods, the optical parameters of the terahertz time-domain spectrum are extracted. The results show that different types of wood have time delay line and amplitude difference in time domain spectrum, the attenuation trend and amplitude are different in frequency domain spectrum, the bands of various types of redwood absorption peaks appear differently in the absorption coefficient spectrum, which all can show the differences between various types of wood, indicating that THz-TDS has feasibility for classification of redwood. The successive projections algorithm (SPA) is used to extract the characteristic frequency of the absorption coefficient spectrum and the refractive index spectrum. 28 characteristic frequency points are selected from the 260 frequency points of the absorption coefficient spectrum and the frequency band accounts for 1077%; 12 characteristic frequency points are selected from 260 frequencies of the refractive index spectrum, and the frequency band accounts for 462%. A random forest classification model and a support vector classification model based on the absorption coefficient spectrum and the refractive index spectrum are established and compared. The results show that THz-TDS has great quality to recognize wood. A random forest classification model based on absorption coefficient spectrum and refractive index spectrum shows good classification performance for redwood species and the accuracy rate of classification is 94% and 96% which can show that they can classify and identify redwood species correctly. THz-TDS technique is used to classify and identify mahogany, which provides a new idea and technical scheme for the classification and identification of mahogany therefore it can be used as a supplement to the near-infrared spectrum wood detection method. This method also provides a theoretical basis to apply terahertz technology in the field of wood classification and identification.