基于PCA-K均值聚类算法的古代玻璃制品的成分分析与鉴别

Component Analysis and Identification of Ancient Glassware Based on PCA-K Mean Clustering Algorithm

摘要: 本文主要对古代玻璃制品相关检测数据进行风化前后规律的研究，根据风化规律进行亚分类，并建立玻璃制品各项化学成分含量风化前的预测模型，同时根据分类规律，对未知类别的玻璃制品进行鉴别。首先采用数据可视化和Pearson卡方检验方法分析表面风化情况与玻璃类型、颜色和纹饰之间的相关性。其次，采用Python软件计算玻璃在风化前后各项化学含量的均值和方差，获得玻璃的风化规律，并利用风化前后各项化学成分含量的平均值的改变比率来预测风化前的化学含量。最后，利用数据可视化分析高钾玻璃和铅钡玻璃的分类规律。采用主成分分析(PCA)和K-Means聚类方法对两种玻璃进行亚分类，并利用ROC曲线对分类模型进行了合理性和敏感性分析。通过分类模型对未知类别的玻璃制品化学成分含量进行分析并鉴别其所属类型，即文物编号A1，A5，A6，A7为高钾玻璃，A2，A3，A4，A8为铅钡玻璃。通过模型敏感性分析，说明分类结果有较好的稳定性和准确性。

Abstract: This paper mainly studies the pre- and post-weathering patterns of the relevant detection data of ancient glassware, conducts subclassification based on the weathering patterns, and establishes a prediction model for the content of various chemical components of glassware before weathering. At the same time, it identifies glassware of unknown categories according to the classification patterns. Firstly, data visualization and Pearson chi-square test methods were adopted to analyze the correlation between surface weathering and the type, color and pattern of glass. Secondly, Python software is used to calculate the mean and variance of each chemical content of the glass before and after weathering, obtaining the weathering law of the glass. Then, the change rate of the average value of each chemical component content before and after weathering is utilized to predict the chemical content before weathering. Finally, data visualization is utilized to analyze the classification patterns of high-potassium glass and lead-barium glass. The principal component analysis and K-Means clustering methods were adopted to subclassify the two types of glass, and the ROC curve was used to analyze the rationality and sensitivity of the classification model. Through the classification model, the chemical composition content of glass products of unknown categories was analyzed and their types were identified. Specifically, the cultural relic numbers A1, A5, A6, and A7 were high-potassium glass, while A2, A3, A4, and A8 were lead-barium glass. Through model sensitivity analysis, it is indicated that the classification results have good stability and accuracy.