基于可解释性的多模型融合的古代玻璃成分分析及亚分类方法  

Ancient Glass Composition Analysis and Sub-classification Methods Based on Interpretable Multi-model Fusion

作　　者：汤思远 黎恒 邱诗睿 朱宇坤 柯圆圆 TANG Siyuan;LI Heng;QIU Shirui;ZHU Yukun;KE Yuanyuan(School of Artificial Intelligence,Jianghan University,Wuhan 430056,Hubei,China;School of Intelligent Manufacturing,Jianghan University,Wuhan 430056,Hubei,China;School of Electronic and Information Engineering,South China University of Technology,Guangzhou 510640,Guangdong,China)

机构地区：[1]江汉大学人工智能学院,湖北武汉430056 [2]江汉大学智能制造学院,湖北武汉430056 [3]华南理工大学电子与信息学院,广东广州510640

出　　处：《江汉大学学报（自然科学版）》2024年第6期37-51,共15页Journal of Jianghan University：Natural Science Edition

基　　金：江汉大学校级科研计划项目(2023JCYJ10);江汉大学2024年研究生科研创新基金项目(KY-CXJJ202441)。

摘　　要：因埋藏环境的影响会使玻璃的内部元素和环境元素进行交换而风化,导致成分比例都发生变化,给考古工作带来一定影响。从玻璃文物的化学成分含量角度进行分析,利用单变量因素分析,卡方检验和SHAP-SVC相融合的方法得到如下结果:与古玻璃风化程度相关的3个指标排序为玻璃类型>纹饰>颜色。通过数据挖掘,将玻璃的类型作为分类变量,将玻璃的化学成分的含量作为呈现变量,对14种化学成分含量进行可视化分析,得出玻璃表面有无风化化学成分含量的统计规律,即当高钾玻璃的SiO_(2)含量高于90%时大概率出现风化现象,铅钡玻璃的SiO_(2)含量低于30%时大概率出现风化现象。然后,利用风化前后的各个成分含量的中位数的差值构建风化预测模型,通过GMM和决策树算法,建立了玻璃的粗分类和亚分类模型,给出了铅钡玻璃亚类主要通过PbO、SiO_(2)、SrO、BaO和CaO这5种化学成分含量进行划分,而高钾玻璃亚类则是通过CaO、Al_(2)O_(3)和SiO_(2)这3种化学成分含量进行划分。Due to the influence of the burial environment,the ancient glass will weather because of its internal elements exchanging with the environmental elements,resulting in changes in the composition proportion,which will have a certain impact on archaeological work.In this paper,we analyzed the chemical compositions of glass artifacts from the perspective of content,using univariate factor analysis,chi-square test,and SHAP-SVC fusion method to analyze the three indicators related to the weathering degree of ancient glass,which were ranked as glass type>decoration>color.Taking the type of glass as a categorical variable and the chemical composition content of glass as a presenting variable,the contents of 14 chemical compositions were visualized and analyzed by data mining,and the statistical laws of the chemical composition content with and without weathering on the glass surface were obtained;that was,when the SiO_(2)content of high potassium glass was higher than 90%,the weathering phenomenon was likely to happen,and when the SiO_(2)content of lead-barium glass was lower than 30%,the weathering phenomenon was expected to happen.Then,the weathering prediction model was constructed using the difference in the median content of each component before and after weathering.The coarse classification and sub-classification model of glass was established based on GMM and the decision tree algorithm.It was given that the lead-barium glass subclasses were mainly divided by the content of PbO,SiO_(2),SrO,BaO,and CaO.In contrast,the high potassium glass subclasses were divided by the content of CaO,Al_(2)O_(3),and SiO_(2).

关 键 词：古代玻璃 玻璃风化 风化预测 玻璃分类 SHAP-SVC 决策树 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TQ171.1[自动化与计算机技术—控制科学与工程]