Analytical identification model of ancient glass based on density clustering and PSO-SVM

Abstract

In this paper, the correlation between the indicators of glass artifacts is explored in depth by establishing relevant mathematical models, giving the statistical laws of weathering for high potassium glass and lead-barium glass, and classifying a series of glass artifacts. In this paper, a support vector machine (PSO-SVM) model based on particle swarm optimization is developed. Firstly, the SVM model is proposed to achieve binary classification for the model, while the particle swarm model is established, and the hyperparameters c and g are optimized by adding adaptive particle variation operation and velocity contraction factor, and the optimal values of both are 0.248 and 2.244, respectively. and the training and validation sets are continuously changed by K-fold cross-validation, and the accuracy of the model classification is obtained as 100%. The model was then used to classify the Form 3 data, and the results were that the artifacts in groups 2, 3, 4, 5, and 8 were lead-barium glass; the artifacts in groups 1, 6, and 7 were high-potassium glass. Finally, the sensitivity analysis of the model was carried out to adjust the values of hyperparameters c and g. It was found that the model sensitivity was very low and the results were satisfactory. In addition, the spearman correlation coefficients between the individual chemical compositions of the two types of glasses were calculated, plotted thermodynamically and interpreted. Then, a gray correlation analysis model was developed to investigate the relationship between other chemical components and it by using silica as the parent sequence, and conclusions were drawn. And based on the results, the differences in the correlation between the chemical components of the two types of glasses were analyzed, and it was found that the correlation between some components in the two types of glasses and the silica content were significantly different.