Component Analysis of Ancient Glass Based on Neural Network

Abstract

The weathering of glass relics is easily affected by the environment. In order to protect cultural relics, we have reduced the number of samples per cultural relic and the number of cultural relics sampled, resulting in limited and incomplete data. These incomplete data are used to predict whether the cultural relics have weathered and reveal the weathering law. It is of practical significance to study the weathering process of glass by analyzing the chemical components of glass. In this paper, the coefficient of variation method is used to extract the chemical components with high contribution rate to weathering and the missing values of each group of data are filled by hot card filling to obtain a set of sample data. Based on the known sample data, we fit the density function of random variables by kernel density and expand the sample size by random number function to design the input data of neural network. The expanded sample value serves as the training set of the neural network and the known sample set serves as the validation set of the neural network. Finally neural network is used for determination of weathering type of sampling points. The neural network designed in this paper realizes the high-precision estimation of weathering types of sampling points.