基于太赫兹时域光谱技术的烟草组分识别

为了准确识别不同烟草配方组分，利用太赫兹时域光谱技术，针对烟草工业常用的叶丝、梗丝和再造烟叶丝3种烟草配方组分开展太赫兹光谱特性分析和分类识别方法研究。对0.35～1.50 THz范围内3种烟丝的吸收系数谱和折射率谱进行分析，通过低方差滤波结合主成分分析（Principal Component Analysis，PCA）进行光谱特征提取和降维，分别建立针对吸收谱和折射谱的支持向量机（Support Vector Machine，SVM）分类模型、最邻近分类（K-Nearest Neighbor，KNN）模型和袋装树（Bagged trees）分类模型。结果表明，基于吸收系数谱的分类模型准确率最高，低方差滤波结合PCA的特征提取算法能显著提高分类效果，其中KNN模型准确率达到98.3%。对频域光谱使用连续投影算法（Successive Projections Algorithm，SPA）特征提取并结合SVM模型，分类准确率也在90%左右。研究表明太赫兹时域光谱技术可应用于不同烟草组分的分类判别，为太赫兹技术在烟草物料无损检测的应用提供参考。

Identification of tobacco materials based on terahertz time-domain spectroscopy

Abstract: Tobacco leaves can often be classified in detail by variety, stalk position, and place of production. Different tobacco components include leaf shreds, stem shreds, and reconstituted tobacco leaf shreds. It is a high demand to classify and identify the tobacco components in recent years. This present work aims to analyze the absorption coefficient spectrum and refractive index spectrum of three tobacco components in the range of 0.35-1.50 THz using the terahertz time-domain spectroscopy. The low-variance filtering combined with Principal Component Analysis (PCA) was performed for the spectral feature extraction and dimension reduction on spectroscopy data. Three classification models were developed to determine the specific absorption and refraction spectra of tobacco, including Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Bagged trees. The results show that a higher accuracy was achieved in the classification model using the absorption coefficient spectrum. The low variance filter combined with the PCA feature extraction significantly improved the classification accuracy, and the KNN model presented an accuracy rate of 98.3%. Furthermore, the Successive Projections Algorithm (SPA) feature extraction was also utilized for the frequency domain spectrum combined with the SVM model, where the classification accuracy was also about 90%. Consequently, the terahertz time-domain spectroscopy technology can be expected to serve the classification of cut tobacco. The finding can provide a strong reference for the application of terahertz time-domain spectroscopy to the non-destructive detection of tobacco materials.