能量色散X射线荧光（Energy dispersive X-Ray Fluorescence，EDXRF）分析技术因其快速、准确、无损等优点，已成为无机元素分析领域最重要的分析技术之一，被广泛应用于地质、环境、材料、考古等领域。在面向古陶瓷类文物的分析时，准确获取其元素组成是进行陶瓷原料来源和烧制工艺研究的基础。然而，古陶瓷类样品主要由Na₂O、MgO、Al₂O₃和SiO₂等轻元素氧化物和其他熔剂组成。使用EDXRF对古陶瓷类样品进行定量分析时，受到诸如基体效应，轻元素自身X射线吸收截面低、荧光产额低，样品及检测路径对特征X射线的强烈吸收等因素的影响，轻元素特征X射线的测量存在一定的困难。特别是对于部分大尺寸和不可移动的古陶瓷文物，受限于真空室尺寸和实际测量条件，其需要在大气环境中进行EDXRF分析，这对轻元素的定量分析方法提出了新的要求。此外，由于古陶瓷类样品各色釉彩分布的不均匀性，还需使用由EDXRF分析技术与X射线光学技术相结合所发展而来的微束能量色散X射线荧光（Micro energy dispersive X-Ray Fluorescence，Micro-EDXRF）分析技术，对古陶瓷类样品进行微区和原位的定量分析。

针对EDXRF分析技术在大气环境中对轻元素的定量分析难题，本文利用机器学习算法对非线性问题求解的显著优势，通过对算法结构的设计，进行光谱数据与各元素浓度间的函数关系的探索。对于Micro-EDXRF分析，本文通过测量毛细管X光透镜传输效率，准确获取了入射X射线的能量分布，并建立了全元素理论α系数库，为常规元素的定量分析提供基础。总的说来，本论文从硬件和软件两方面出发，对EDXRF的定量分析方法开展系统研究。本文主要的研究内容和结论如下：

（1）基于反向传播神经网络（Backpropagation Neural Network，BPNN）算法开发了一个适用于EDXRF在大气环境中对轻元素进行定量分析的模型。使用EDXRF谱仪对土壤、岩石标准样品以及由它们和化合物所烧制的陶瓷标准样品进行光谱测量，并将该光谱数据作为模型训练数据的集合。所设计的模型使用经预处理的康普顿散射数据和与轻元素具有相关性的其他元素浓度作为输入数据进行训练。测试结果表明，该模型对O、Na、Mg和Al元素的预测浓度与真实浓度间具有良好的一致性，能够满足轻元素定量分析的需要。此外，面向常规元素的定量分析，本文建立了一种改进型的基本算法。文中通过计算两组标样中不同元素在纯元素条件下的特征X射线强度的，验证了该定量分析方法对不同类型样品的适应性。

（2）基于Python语言环境开发了一款EDXRF定量分析软件QAXRF。该软件使用PyQt函数库进行软件界面的编写，其能实现与光谱数据预处理、理论α系数计算及定量分析等相关的各项功能。通过对比分析，QAXRF与欧洲同步辐射光源所开发的Pymca软件对常规元素的定量分析结果相近。此外，由于QAXRF使用了改进的基本算法（Fundamental Algorithm，FA）对常规元素进行定量分析，其可以通过导入毛细管X光透镜的能量传输效率实现Micro-EDXRF的定量分析。同时，QAXRF可通过轻元素定量分析模型的导入，利用EDXRF光谱中的Compton散射数据和相关元素含量，实现样品中轻元素（Na、Mg、Al等）的准确定量。

（3）利用所开发的EDXRF定量分析方法，开展其对古陶瓷类样品的应用研究。为探索古代铅釉陶瓷烧制工艺的变化及其发展脉络，分别对来自不同窑址的唐三彩残片和来自不同朝代的西岳庙古琉璃残片进行无损分析。实验结果为铜绿釉和铁黄釉在历朝历代的发展过程的延续性提供佐证。此外，本文提出了一种基于X射线透射法测量陶瓷样品真气孔率的方法。该方法将陶瓷样品的透射光谱与EDXRF分析结果相结合，通过计算陶瓷样品在不同能量条件下的理论线衰减系数与测量线衰减系数的比值，实现了古陶瓷样品真气孔率的准确测量。

综上所述，本文面向EDXRF在大气环境中进行元素定量分析时所面临的困难，从X射线荧光定量分析的基本原理和机器学习的非线性建模两部分出发，开发适用于EDXRF分析技术的定量分析方法。此外，本文建立适用于X射线荧光元素定量分析方法的完善软硬件体系，为不同类别样品的分析需求提供了实验条件。由此可见，该定量分析方法具有广泛的应用前景。

Energy dispersive X-ray fluorescence (EDXRF) analysis technology has become one of the most important analytical technologies in the field of inorganic element analysis due to its advantages of being fast, accurate, and non-destructive. So, it was widely used in geology, environment, materials, archaeology, and other fields. Accurately obtaining the elemental composition of ancient ceramic samples is the basis for studying the sources of ceramic raw materials and firing techniques. However, ancient ceramic samples are mainly composed of light element oxides such as Na₂O, MgO, Al₂O₃, and SiO₂, as well as other fluxes. When using EDXRF for quantitative analysis of ancient ceramic samples, it is hard to measure the characteristic X-rays of light elements due to factors such as matrix effect, low X-ray absorption cross-section and low fluorescence yield of light elements, and strong absorption by the sample itself and the detection path. Especially for some large-sized or immovable ancient ceramic relics, limited by the size of the vacuum chamber and actual measurement conditions, EDXRF analysis needs to be conducted in the atmospheric environment, which poses new requirements for quantitative analysis methods of light elements. In addition, due to the uneven distribution of various glaze colors in ancient ceramic samples, it is necessary to use micro energy dispersive X-ray fluorescence (Micro-EDXRF) analysis technology developed by combining EDXRF analysis technology with X-ray optical technology to quantitatively analyze ancient ceramic samples in micro and situ.

In response to the challenge of quantitative analysis of light elements in the atmospheric environment using EDXRF analysis technology, this paper utilizes the significant advantages of machine learning algorithms in solving nonlinear problems. By designing the algorithm structure, we explore the functional relationship between spectral data and the concentrations of various elements. For Micro EDXRF analysis, this paper accurately obtained the energy distribution of incident X-rays by measuring the transmission efficiency of polycapillary X-ray optics. And it establishes the all-element theory α coefficient library and provides a foundation for quantitative analysis of conventional elements. Overall, this paper conducts a systematic study of the quantitative analysis method of EDXRF from both hardware and software perspectives. The main research content and conclusions of this article are as follows:

(1) A model suitable for EDXRF quantitative analysis of light elements in atmospheric environments has been developed based on the Backpropagation Neural Network (BPNN) algorithm. An EDXRF spectrometer was used to perform spectral measurements on soil and rock standard samples, and ceramic standard samples fired from the standard samples and compounds. And these spectral data were used as a set of model training data. The model used the preprocessed Compton scattering data and other element concentrations that are correlated with light elements as input data for training. The test results indicate that this model has good consistency between the predicted and true concentrations of O, Na, Mg, and Al elements. So, this model can meet the needs of quantitative analysis of light elements. In addition, this paper established an improved fundamental algorithm for quantitative analysis of conventional elements that can be measured by characteristic X-ray. The adaptability of the quantitative analysis method to different types of samples was verified by calculating the characteristic X-ray intensity of each element in the two groups of standard samples under pure element conditions.

(2) Based on the Python language environment, an EDXRF quantitative analysis software was developed. This software uses the PyQt function library for software interface writing, which can achieve spectral data preprocessing and theoretical analysis α Various functions related to the coefficient calculation and quantitative analysis. Through comparative analysis, the quantitative analysis results of conventional elements by QAXRF and Pymca software developed by European synchrotron radiation Light Source are similar. In addition, since QAXRF uses an improved Fundamental algorithm for quantitative analysis of conventional elements, it can achieve a quantitative analysis of Micro-EDXRF by introducing the energy transfer efficiency of capillary X-ray lenses. At the same time, QAXRF can accurately quantify light elements (Na, Mg, Al, etc.) in the sample by importing a quantitative analysis model of light elements and utilizing Compton scattering data from EDXRF spectra and related element concentration.

(3) The developed EDXRF quantitative analysis method was used to conduct research on its application to ancient ceramic samples. To explore the changes and development of ancient lead-glazed ceramic firing technology, non-destructive analysis was conducted on Tang Sancai fragments from different kiln sites and Xiyue Temple ancient glazed fragments from different dynasties. The experimental results provide evidence for the continuity of the development process of copper green glaze and iron yellow glaze in various dynasties. In addition, this paper proposes a method for measuring the true porosity of ceramic samples based on the X-ray transmission method. This method combines the transmission spectrum of ceramic samples with the results of EDXRF analysis. By calculating the ratio of the theoretical linear attenuation coefficient and measured linear attenuation coefficient of ceramic samples under different energy conditions, the accurate measurement of the true porosity of ancient ceramic samples is realized.

In summary, this paper aims at the difficulties faced by EDXRF elemental quantitative analysis in the atmospheric environment. It develops a quantitative analysis method suitable for EDXRF from the basic principles of X-ray fluorescence quantitative analysis and the nonlinear modeling of machine learning. In addition, this paper establishes a complete software and hardware system suitable for the quantitative analysis of X-ray fluorescence light elements, which provides experimental conditions for the analysis requirements of different types of samples. So, the quantitative analysis method has broad application prospects.