人工神经网络与多元统计判别分析在古陶瓷断源断代中的对比研究

Comparison between Artificial Neural Networks and Multivariate Statistical Discriminant Analysis Applied to Ancient Ceramic Provenance and Chronology

针对古陶瓷断源断代的研究目的,以景德镇仿龙泉青瓷与龙泉青瓷胎的化学组成为研究对象,选择了多元统计判别分析方法、人工神经网络算法对其进行数据处理分析和产地判别研究,并将多元统计判别分析和人工神经网络进行对比分析,探讨了不同方法之间的差异性及适用性。结果表明,由于古陶瓷元素组成数据难以完全满足多元统计判别分析对于数据变量的要求,因而多元统计判别分析相对于人工神经网络的判别正确率较低,人工神经网络更适合用于古陶瓷断源断代研究。

The chemical composition test results of the body of Jingdezhen imitated Longquan and Longquan celadon were studied by multivariate statistical discriminant analysis and artiifcial neural networks for their respective provenance and chronology. The differences and applicability of the two methods were discussed. Results show that as the data of ancient ceramic element composition couldn’t fully meet its requirements, the accuracy of the multivariate statistical discriminant analysis is lower than that of the artiifcial neural networks, which means the artiifcial neural networks is more suitable for ancient ceramic provenance and chronology determination.