【作者】 陈涛；

【导师】 莫玮； 李智；

【作者基本信息】 西安电子科技大学 ， 测试计量技术及仪器， 2013， 博士

【摘要】 太赫兹（THz）波是目前国际学术界公认的一个非常重要的前沿热点领域。由于THz波具有较强的光谱分辨本领以及良好的透视性和安全性，THz技术被认为在物质检测方面具有广阔的应用前景。太赫兹时域光谱（THz-TDS）技术是基于飞秒超快激光技术的THz波段光谱测量新技术，它是目前应用THz波研究物质的主要技术之一。本文对基于THz-TDS技术的物质定性鉴别和定量分析方法进行了系统研究，所取得的主要研究成果为：1.对THz-TDS技术在安全检查领域中的爆炸物探测和识别方法进行了研究。提出了一种基于模糊聚类分析的模糊模式识别方法，成功应用于爆炸物的THz光谱分类和识别中。通过对多种单质炸药和混合炸药及几种可能与它们混淆的物质的THz特征吸收光谱进行研究，提取它们在0.2~2.2THz频率范围内4个相对较强的特征吸收峰作为分类识别的特征参数。首先，用基于模糊等价关系的模糊聚类分析方法对这几种物质的THz特征吸收谱进行聚类训练，获得样品分类并形成标准THz吸收光谱模型库；然后，选用与训练样本主体成分相同的两种物品作为待识别对象，采用基于择近原则的模糊模式识别方法进行了成功识别。研究结果表明，基于模糊聚类分析的模糊模式识别方法对THz吸收谱具有较强的相似特征聚类功能和较高的识别率，为THz光谱技术用于爆炸物的检测和识别提供了一种有效的方法。2.对THz-TDS技术在无损检测和质量控制领域中的物质定性鉴别方法进行了研究。首次尝试将主成分分析（PCA）和模糊模式识别相结合的方法应用于生物分子的THz光谱识别中，并以一些典型氨基酸和糖类生物分子的THz透射谱作为实验样本证明了此方法的可行性。首先，应用主成分分析法对原始光谱变量进行降维处理，提取生物分子样品的THz光谱特征信息，获得样品分类并形成标准THz光谱模型库；然后，用选取的主成分得分矩阵代替原始光谱变量输入模糊模式识别模型中，采用基于择近原则的模糊识别方法对样品进行了成功识别。研究结果表明，以生物分子的THz光谱作为数据特征，采用主成分分析与模糊模式识别相结合的方法实现生物分子的识别和鉴定是可行的，为提高生物分子的THz光谱识别准确率和实现生物分子THz光谱的自动识别提供了一种新的有效的分析方法，在食品安全、药品安全及药物质量控制和成分分析等领域具有重要的现实意义和应用价值。3.对THz-TDS技术在无损检测和质量控制领域中的物质定量分析方法进行了研究。提出了一种应用THz-TDS技术结合化学计量学方法对多组分药物混合物中药物活性成分（API）和药用辅料的含量进行同时定量测定的方法，并以无水茶碱、乳糖一水合物和硬脂酸镁三元药物混合物75个样品及对乙酰氨基酚、乳糖一水合物、微晶纤维素和可溶性淀粉四元药物混合物100个样品作为实验介质验证了所提方法的可行性和有效性。实验采用THz-TDS系统测量了无水茶碱、乳糖一水合物和硬脂酸镁三元药物混合物及对乙酰氨基酚、乳糖一水合物、微晶纤维素和可溶性淀粉四元药物混合物的THz吸收光谱，并采用主成分回归（PCR）和偏最小二乘回归（PLS）两种多元校正方法分别建立了THz吸收谱与多元药物混合物各组分含量的定量回归模型，同时获得了混合物中药物活性成分和药用辅料的含量。PLS回归取得更好结果，其中三元混合物两组分含量的PLS定量模型校正及预测相关系数R均高于0.98，四元混合物中对乙酰氨基酚、乳糖一水合物、微晶纤维素和可溶性淀粉含量的PLS定量模型校正及预测相关系数R均分别高于0.93，0.98，0.63和0.86。研究结果表明，THz-TDS技术结合化学计量学方法建立定量分析模型能够实现对多元混合物成分含量的无损非破坏定量分析，在药物分析等领域将有广阔的应用价值。4.对多种特征谱区筛选算法进行了深入研究，并成功引入到多元混合物的THz光谱定量分析中，有效提高了多元混合物THz光谱定量分析模型的精度和降低了模型复杂度。实验利用THz-TDS系统测量了乳糖一水合物、对乙酰氨基酚、微晶纤维素和可溶性淀粉四元药物混合物的THz吸收光谱，并分别尝试采用常规区间偏最小二乘（iPLS）、向后区间偏最小二乘（biPLS）、联合区间偏最小二乘（siPLS）和移动窗口偏最小二乘（mwPLS）谱区筛选方法对多元混合物的THz吸收光谱进行特征子区间优选，建立了THz吸收谱与四元混合物中乳糖一水合物含量之间的定量回归模型。不同谱区筛选算法所得的结果表明，mwPLS谱区筛选模型得到的结果相对最优，其校正集和预测集相关系数R分别为0.9960和0.9951，交互验证均方根误差（RMSECV）和预测均方根误差（RMSEP）分别为0.9803和1.1141。研究结果表明，采用特征谱区筛选方法可以有效选择多元混合物THz吸收光谱的特征区间，剔除多元混合物THz吸收光谱中不相关或非线性变量的信息，提高模型精度和降低模型复杂性，实现对多元混合物成分含量快速、高效、无损的定量检测，在生物、食品和药物分析等领域将有广阔的发展前景和应用价值。更多还原

【Abstract】 In recent years, terahertz (THz) wave has become an extremely attractive researchfield. Since THz wave exhibits the properties of spectroscopy, good penetration andsafety, THz technology has promising applications for the detection and identification ofmaterials. Terahertz time-domain spectroscopy (THz-TDS) is a new technique thatapplied to spectroscopic measurement based on ultrafast femtosecond laser, and it iscurrently a key technique for the study of materials using THz radiation. Thisdissertation is focusd on developing new methods for qualitative and quantitativedetection of materials in the fields of security inspection, nondestructive testing (NDT)and quality control (QC) based on THz-TDS. The author’s major contributions areoutlined as follows:1. Methods for detection and identification of explosives in the field of securityinspection based on THz-TDS have been studied. An approach for automaticidentification of THz spectra of explosives is proposed based on fuzzy patternrecognition with fuzzy cluster analysis. On the basis of studying the absorption spectraof several pure explosives, mixed explosives and potential confusion materials，fourrelatively strong characteristic absorption peaks in the frequency range from0.2to2.2THz are extracted as the characteristic parameters of classification and recognition, andfuzzy pattern recognition is used to carry out the classification and recognition of theirTHz characteristic absorption spectra. Firstly, fuzzy cluster analysis method based onfuzzy equivalent matrix is used to cluster the THz absorption spectra of these materials,thus the sample clustering is obtained and the standard model library of THz absorptionspectra is formed. Secondly, two materials, which have the same main ingredient as thatof the training samples, are used as the objects to be identified; and the fuzzy patternrecognition method based on the principle of fuzzy closeness optimization is adopted toidentify the objects. Study results indicate that the fuzzy pattern recognition methodbased on fuzzy cluster analysis has strong similar character clustering function and highrecognition rate for THz absorption spectra, which provides an effective method in thedetection and identification of explosives using THz spectroscopy.2. Methods for qualitative detection of materials in the fields of nondestructivetesting and quality control based on THz-TDS have been studied. We report the first useof principal component analysis (PCA) combined with fuzzy pattern recognition toidentify the THz spectra of biomolecules in this dissertation, and THz transmittancespectra of some typical amino acid and saccharide biomolecular samples are investigated to prove its feasibility. Firstly, PCA is applied to reduce the dimensionalityof the original spectrum data and extract features of the data. Secondly, instead of theoriginal spectrum variables, the selected principal component scores matrix is fed intothe model of fuzzy pattern recognition, where a principle of fuzzy closeness basedoptimization is employed to identify those samples. Study results demonstrate that THzspectroscopy combined with PCA and fuzzy pattern recognition can be efficientlyutilized for automatic identification of biomolecules. The proposed approach provides anew effective method in the detection and identification of biomolecules using THzspectroscopy.3. Methods for quantitative determination of materials in the fields ofnondestructive testing and quality control based on THz-TDS have been studied. Anapproach for simultaneous quantitative determination of both active pharmaceuticalingredient (API) and excipient concentrations of multicomponent pharmaceuticalmixtures is proposed using THz-TDS with chemometrics, and75ternary mixturesformulated with anhydrous theophylline, lactose monohydrate, magnesium stearate and100quaternary mixtures composed of acetaminophen, lactose monohydrate,microcrystalline cellulose and soluble starch are investigated to prove its feasibility andefficiency. The THz spectra for ternary mixtures of anhydrous theophylline, lactosemonohydrate, magnesium stearate, and quaternary mixtures of acetaminophen, lactosemonohydrate, microcrystalline cellulose and soluble starch are measured usingTHz-TDS. Two multivariate calibration methods, principal component regression (PCR)and partial least squares (PLS) regression, are employed to correlate THz absorbancespectra with the pharmaceutical tablet concentrations. Both API and excipientconcentrations of mixtures are predicted simultaneously, and the PLS method providesbetter result than PCR method. The correlation coefficients of calibration (Rcal) andvalidation (Rval) for ternary mixtures’ components, anhydrous theophylline and lactosemonohydrate, are all more than0.98. The Rcaland Rvalfor quaternary mixtures’components, acetaminophen, lactose monohydrate, microcrystalline cellulose andsoluble starch, are all more than0.93,0.98,0.63and0.86, respectively. Experimentalresults show that THz-TDS combined with chemometrics is feasible in nondestructivequantitative analysis of multicomponent mixtures, and it can be widely applied in thefields of pharmaceutical analysis and others.4. Interval selection algorithms are investigated systematically in this dissertation,and THz-TDS coupled with these algorithms is used to perform quantitative analysis of component concentrations in multicomponent mixtures, which can effectively improvethe precision and reduce the complexity for the quantitative analysis models. The THzspectra of100quaternary pharmaceutical mixtures composed of lactose monohydrate,acetaminophen, microcrystalline cellulose and soluble starch are measured usingTHz-TDS system. Four spectral interval selection methods, including interval partialleast-squares regression (iPLS), backward interval PLS (biPLS), synergy interval PLS(siPLS) and moving window PLS (mwPLS), are employed to select spectral intervals ofTHz absorbance spectra of multicomponent mixtures and correlate THz absorbancespectra with the concentrations of lactose monohydrate. Compared to the other threeinterval selection methods and full-spectrum PLS, the mwPLS method yields the mostaccurate result. The optimal mwPLS model is obtained with higher correlationcoefficient for calibration (RC) of0.9960, higher correlation coefficient for prediction(RP) of0.9951, lower root mean square error of cross-validation (RMSECV) of0.9803,and lower root mean square error of prediction (RMSEP) of1.1141. Study resultsindicate that spectral interval selection combined with THz-TDS could be successfullyapplied as an accurate and rapid method to determine component concentrations inmulticomponent mixtures, and it has potential application in the fields of biology, food,pharmaceutical analysis and so on.更多还原