基于原始信号校正的高性能近红外光谱分析新方法研究

Associated with chemometrics, near infrared (NIR) spectroscopy has become a powerful technique in many fields due to its great efficiency and non-destructivity analytical capability. At present, the near infrared spectral analysis methods have the following deficiencies:(1) Interferogram phase correction theory is not systematic; (2) Recognition ability of abnormal sample is weak; (3) The performance of a single multivariate calibration algorithm is inadequate; (4) Evaluation theory for prediction the performance of multivariate calibration algorithms is inadequate. These deficiencies have a significant effect on the performance of quantitative analysis. The purpose of present project is to improve the performance of quantitative analysis for NIR analysis based on the research methods of information science. Series methods will be studied. Firstly, interferogram will be corrected by parametric and non-parametric model, and the high quality spectra are calculated from interferogram which is the original information obtained by Fourier transform spectrometer. Secondly, combined with previous research results, methods for abnormal sample recognition, methods for multivariate calibration considering fusion correction and methods for forward model evaluation will be studied. During these processes, multidimensional signal processing technology, optimization method, and adaptive processing technology will be applied. Thirdly, a systematic theory of near infrared (NIR) spectroscopy analysis will be constructed which can improve the quantitative analytical capability. The research can enrich the theory of near infrared spectra analysis. These research methods can be applied to other spectrum analysis such as ultraviolet spectrum, fluorescence spectrum and chromatography analysis.

项目以提高近红外光谱定量分析性能为目标，针对目前近红外光谱分析方法存在干涉图校正理论不够完善、异常样本识别能力有限、单一多元校正算法优势发挥不足以及算法预测性能评价理论不够完善等问题，融合信息学科研究方法，突破化学学科固有的研究思路限制，从傅里叶变换光谱仪获得的原始信号——干涉图出发，通过参数化模型与非参数化模型研究干涉图校正方法，从而获得高质量的近红外光谱。在此基础上，结合前期多元校正与异常光谱识别研究成果，利用多维信号处理、最优化方法、自适应处理等理论技术，研究与多元校正相关的异常样本识别、融合校正、前向模型评价等科学问题，最终形成一套近红外光谱分析理论体系并实现提高光谱定量分析性能的目标。该项目研究成果能丰富近红外光谱分析理论，成果中的研究方法、思路及某些算法可应用于其它谱（如紫外光谱、荧光光谱、色谱等）分析。

针对目前近红外光谱分析方法存在的不足，融合信息学科研究方法，突破化学学科固有的研究思路限制，从傅里叶变换光谱仪获得的原始信号——干涉图出发，研究近红外光谱分析新方法。项目主要提出了干扰图校正新理论，建立干涉图测量模型，从统计角度对干涉图进行分析，分别建立干涉图相位校正参数化模型与非参数化模型，并通过实验验证了模型的可用性；从信息论角度对光谱数据结构进行分析，突破传统多元校正利用主成分信息的思维，提出利用非主成份信息进行光谱分析的新方法，提高了预测结果的准确性与鲁棒性；针对现有多元校正算法各有优缺点的事实，充分利用各算法的优点克服其缺点，提出基于信息融合的多元校正方法，使各算法的预测误差相互抵消，从而降低总预测误差，通过实验验证了算法的有效性。本项目提出的近红外光谱分析方法既自成理论体系又可将其中部分应用于现有实验数据的处理，这些研究思路与方法还可应用于其它谱（如紫外光谱、荧光光谱、色谱等）分析。