激光探针技术中光谱数据处理方法研究

发布时间：2018-04-10 09:40

本文选题：激光探针　切入点：光谱数据处理　出处：《华中科技大学》2015年硕士论文

【摘要】：自激光探针技术诞生以来,研究者主要集中于激光诱导等离子体的物理特性、实验样品的物理化学性质、实验参数优化和仪器设备性能等方面的研究,光谱数据处理方法的研究没有得到足够重视。然而作为一种有力的软优化方法,数据处理方法有着诸多方面的优势。一方面,数据处理可以代替某些高精度硬件设备实现技术指标,降低硬件成本;同时解决硬件优化无法克服的技术难题,提升光谱分析的质量;另一方面,数据处理通过提取光谱的有用信息并对数据进行再加工,可以显著提高光谱分析的精度。本文通过对光谱数据预处理方法和定量分析方法的研究,有效提高了定性和定量分析的准确度。首先,本文研究了基于连续小波变换的激光探针谱峰识别算法,提出了一种自动计算噪声的新方法,采用连续小波变换结合信噪比阈值法实现谱峰的自动识别。将该方法应用于土壤样品的光谱中,结果表明,该方法能够有效排除尖峰噪声的干扰,识别强峰,并且具有较强的重叠峰分辨能力和良好的定性分析能力,为后续的定量分析奠定了基础。其次,本文提出了基于离散小波变换背景扣除的改良算法,对传统的背景拟合算法进行修正。通过对微合金钢样品的激光探针光谱进行背景校正,并对Cr、V、Cu和Mn元素进行定量分析。结果表明,该方法能够使谱线的背景明显降低,并能有效避免出现背景的高估现象,与未进行背景校正、多项式拟合背景扣除方法和常规的小波变换方法相比,这种方法能够改善光谱质量,提高回归模型的准确性。然后,本文研究了基于遗传算法和偏最小二乘法相结合的定量分析模型。通过对11种土壤组成成分Mn、Cr、Cu、Pb、Ba、Al2O3、CaO、Fe2O3、MgO、Na2O和K2O的含量分别进行预测,证明遗传算法作为谱线选择的一种预处理方法,能够有效去除光谱中重复、多余或不相关的变量,减少用于偏最小二乘法建模的光谱谱线数目,从而减少建模时间,最终简化模型。对于大部分土壤组成成分,该模型都能够显著改善定量分析的准确度。最后,本文研究了基于偏最小二乘法和人工神经网络相结合的定量分析模型。将该模型应用于激光探针土壤定量分析中,对11种土壤组成成分的含量进行预测。结果表明,该模型能够将偏最小二乘法降低自变量多重共线性和人工神经网络具有非线性处理能力的优势结合起来,显著改善了激光探针定量分析的准确度。
[Abstract]:Since the birth of laser probe technology, researchers have focused on the physical properties of laser-induced plasma, the physical and chemical properties of experimental samples, the optimization of experimental parameters and the performance of instruments and equipment.The research of spectral data processing method has not been paid enough attention to.However, as a powerful soft optimization method, data processing method has many advantages.On the one hand, data processing can replace some high-precision hardware devices to achieve technical targets and reduce the cost of hardware; at the same time, it can solve the technical problems that can not be overcome by hardware optimization, and improve the quality of spectral analysis; on the other hand,The accuracy of spectral analysis can be significantly improved by data processing by extracting useful spectral information and reprocessing the data.In this paper, the preprocessing method and quantitative analysis method of spectral data are studied to improve the accuracy of qualitative and quantitative analysis.Firstly, this paper studies the laser probe spectral peak recognition algorithm based on continuous wavelet transform, and proposes a new method to automatically calculate the noise. The continuous wavelet transform combined with the SNR threshold method is used to realize the automatic recognition of the spectral peak.The method is applied to the spectrum of soil samples. The results show that this method can effectively eliminate the interference of peak noise, identify strong peaks, and have strong resolution ability of overlapping peaks and good qualitative analysis ability.It lays a foundation for further quantitative analysis.Secondly, an improved background deduction algorithm based on discrete wavelet transform is proposed to modify the traditional background fitting algorithm.The laser probe spectra of microalloyed steel samples were calibrated and the elements of Cr (V) Cu and mn were quantitatively analyzed.The results show that the proposed method can reduce the background of spectral lines obviously, and can effectively avoid the phenomenon of background overestimation. Compared with the methods of background correction, polynomial fitting background subtraction and conventional wavelet transform, the proposed method can effectively avoid background overestimation.This method can improve the spectral quality and improve the accuracy of the regression model.Then, the quantitative analysis model based on genetic algorithm and partial least square method is studied.The number of spectral lines used for partial least square modeling is reduced, thus the modeling time is reduced and the model is simplified.For most soil components, the model can significantly improve the accuracy of quantitative analysis.Finally, the quantitative analysis model based on partial least square method and artificial neural network is studied.The model was applied to the quantitative analysis of soil with laser probe, and the contents of 11 soil components were predicted.The results show that the model can combine the advantages of the partial least square method to reduce the independent variable multiple collinearity and the artificial neural network has the ability to deal with nonlinear, and improve the accuracy of laser probe quantitative analysis.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP274.2;TN249

【参考文献】