基于主成分分析-BP神经网络的文物识别

Recognition of Cultural Relics Based on Principal Component Analysis and BP Neural Network

摘要: 当下中国文物面临的重大灾难，不是所谓的文物外流，而是文物的卖假。从金缕玉衣到汉代玉凳，一个个的赝品将文物鉴定权威击垮。文物造假为中国带来了文物危机，扭曲了博大精深的传统艺术，故建立合适的文物样品识别模型尤为重要。基于此，本文选取玻璃文物样品作为研究对象，对其化学成分进行数学分析。首先利用加权平均预测法对受损文物的化学成分数据进行修正，并借助斯皮尔曼相关系数对文物样品的化学成分进行相关性分析；其次依据主成分分析数学原理提取出了文物样品的前5个主成分；最后结合BP神经网络，对样本空间进行重构，构建了主成分分析-BP神经网络玻璃文物识别模型，达到了对输入空间降维的目的。经检验，文物模型识别的效果极好。

Abstract: At present, the major disaster facing China’s cultural relics is not the so-called outflow of cultural relics, but the counterfeiting of cultural relics. From golden strands of jade clothes to Han Dynasty jade stools, fakes one by one knocked down the cultural relics identification authority. The falsifi-cation of cultural relics has brought a cultural relic crisis to China and distorted extensive and profound traditional art, so it is particularly important to establish a suitable cultural relic sample identification model. Based on this, this article selects a batch of glass samples. Firstly, the chemical composition data of damaged cultural relics are modified by weighted average prediction method, and the correlation of chemical composition of cultural relic samples is analysed with the help of Spearman correlation coefficient. Secondly, according to the mathematical principle of principal component analysis to extract the sample of first five principal components of cultural relics. Finally, combined with the BP neural network, a principal component analysis (PCA)-BP neural network model for cultural relics recognition is constructed, by reconstructing the sample space to achieve the purpose of dimension reduction in the input space, so as to further simplify the network structure. After testing, the effect of model recognition is excellent, which greatly improves the accuracy of the cultural relic recognition model.