论文标题:野外矿物近红外光谱分析关键技术研究
The Study on Key Techniques for the Field Mineral Near-Infrared Spectroscopy Analysis
论文作者 王智宏
论文导师 林君,论文学位 博士,论文专业 地球探测与信息技术
论文单位 吉林大学,点击次数 94,论文页数 128页File Size1123k
2005-06-01论文网 http://www.lw23.com/lunwen_398586787/ 近红外光谱分析;野外矿物;便携式仪器;数学模型;数据预处理;光谱预处理;矩阵预处理;消噪预处理;应用模型;偏最小二乘;人工神经网络;数据后处理
Near-Infrared Spectroscopy(NIRS) analysis; field minerals; portable instrument; mathematics model; data pretreatment; spectrum pretreatment; matrix pretreatment;den-noise pretreatment; application model; Partial Least Squares(PLS);Artificial Neural Net(ANN); post-treatment
为改变我国便携式野外矿物近红外光谱仪器依赖进口以及国外现有分析软件不适用于我国样品的现状,本文研究并解决野外矿物近红外光谱分析技术的近红外光谱仪器、数据处理、建模方法等三大内容中的关键问题。开发了便携式野外矿物近红外光谱分析仪的光学、机械系统,仪器性能测试及实验对比表明其主要技术指标已接近或达到国外同类产品的水平。针对提出的野外矿物光谱数据的数学模型,系统地归纳了三类光谱数据预处理方法,提出三类方法联合使用的互利原则。采用偏最小二乘法和人工神经网络两种方法,研究建立矿物近红外光谱分析应用模型方法,分析数据的相关性、一次和常数项误差、光谱噪声对模型性能的影响,并提出消除其影响的措施。根据野外矿物半定量分析要求,提出了分析数据的后处理方法,设计了与自制仪器配套的野外矿物成分近红外光谱半定量分析软件并开发了核心模块。
Near-Infrared Spectroscopy (NIRS) analysis technique is a physics method to analyze the matter’s chemistry composition and content by its emission or absorption spectra reflecting its atoms’and molecules’characters. It is a modern analysis technique that integrates multi-subject knowledge: spectroscopy, chemometrics, computer application, etc. The laboratorial instruments are used to measure the mineral composition of rocks or soil in situ by portable NIRS analysis techniques. This is a dream of the field geologic workers for a long time. As that the internal portable field mineral NIRS instruments are importation, and the existing external analysis softwares are not fit for internal mineral samples, combined with Chinese geologic investigation item “The study of the field mineral NIRS analysis method and development of instrument”(item number: 199920190116) and Jilin university innovation fund item “The study of the soil environment Near-Infrared detection information analysis system in situ”, the key technical problems in three aspects of the field mineral NIRS analysis technique--NIR Spectroscopy instrument, Spectroscopy data processing and application model--have been studied and resolved. The main study work and productions: 1. The optical path and machinery system of the portable field mineral NIR spectrum instrument have designed and realized. Its spectrometer is a cross single optical path grating scan framework to settle the conflict between the setting space of parts and the system volume—to supply the enough setting space for the integrating sphere and the modulator, and to ensure the system volume is small. Its machinery includes the precise parts, inching institutions, and a sines gearing to ensure its precise is enough. The capability test and experiment comparison with the congener foreign instrument have done. Its results indicate its main technical parameters have been close to the level of the congener foreign products. 2. The characteristics and processing methods of minerals NIRS data have been researched. The characteristics of minerals NIRS diffuse reflectivity data are that the NIRS data of the same mineral sample have a holistic excursion and range variety, the dynamic range of NIRS data of the same composition minerals is large, the signal-to-noise of different NIRS data are not same. The mathematics model of mineral NIRS data has been concluded. It’s: y = byt +a+n. 3. The three classes of pretreatment methods for spectrum data have been concluded by the numbers, their application effects have been analyzed, and their application principles have been concluded. The spectrum pretreatment can eliminate or decrease the errors brought by spectrum stability and sample characters. The de-noise pretreatments can eliminate or decrease the spectrum noise. The matrix pretreatment can enlarge the differences between the data of the modeling samples, and be used to process the sample data set before modeling. The combining application of the three classes mutual benefit pretreatment approaches can heighten the accuracy and robustness of the model. The application effect of every kind of data pretreatment has relation with its processing data, so the data pretreatment approaches should be ensured after comparing and analyzing all kinds of the approaches. 4. The establishing methods of application model on the mineral NIR analysis have been studied. With Partial Least Squares (PLS) and Artificial Neural Net (ANN) modeling methods, and simulative data, the factors affecting the model performance have been educed. Those are the relativity, the errors of constant and simple part, and spectrum noise of the modeling data. The methods to avoid those factors have been brought forward, their applying impacts have been studied, and their accuracy and robustness have been compared. Conclusions drew from the two modeling methods are same by and large. The relativity of the modeling data affects the accuracy of the model. The errors of constant and simple part, and spectrum noise of the spectrum data depr
|
