玻璃文物的类别鉴定与亚类划分

Classification Identification and Sub-Classification of Glass Relics

摘要: 基于已知玻璃类型的数据，研究玻璃文物表面风化的相关因素和由风化引起的主要化学成分的变化，由 此基于主要化学成分建立玻璃的分类模型和亚类划分模型，并利用所得模型对未知类型的玻璃文物进行 鉴别。研究发现：玻璃表面风化与其类型相关，与纹饰弱相关，而与颜色基本无关；与高钾玻璃相比， 铅钡玻璃表面更易受风化影响；表面风化会引起高钾和铅钡玻璃一些成分比例的显著变化；精选4个重 要成分PbO、BaO、SiO₂、K₂O，建立玻璃类型的随机森林(RF)和支持向量机(SVM)模型，分类准确率均 为100%，使用这些模型对8个未知类型的玻璃文物进行鉴别，结果一致，均为：A1 A6 A7为高钾，其余 为铅钡；对每个类型的玻璃各挑选8个重要成分，使用k-means聚类算法对每个类型玻璃可以很好地细 分为两个亚类；最后，通过对成分数据增加微小随机扰动来证实所建模型是稳健的。

Abstract: Based on glass relics data with known glass types, factors and important chemical compositions related to the weathering of glass surface are studied. And classification models of glass types and sub-classification models are established based on their important chemical compositions. these classification models are applied to identify glass relics with unknown types. Results show that the weathering of glass surface is related to its type, weakly related to its decoration, and unrelated to its color. The surface of PbO-BaO glass is more susceptible to weathering than that of high K₂O glass. Surface weathering can cause significant changes in the proportion of some compositions of high K₂O and PbO-BaO glass. Four important compositions, PbO, BaO, SiO₂ and K₂O, are selected to make models of glass types via random forest (RF) and support vector machine (SVM), with classification accuracy of 100%. These models are employed to identify 8 unknown types of glass relics, same results are presented that A1, A6 and A7 are high K₂O glass, and the rest are PbO-BaO glass. Then 8 important compositions are selected for each glass type, both of which are divided into two subclasses via k-means clustering very well. Finally, the robustness of the proposed models is verified by adding a small random disturbance to the composition data.