Mining Painted Cultural Relics Implied Information Based On Dimension Reduction Of Hyperspectral Images And Image Fusion

Painted cultural relics have abundant variety and color.They are an important part of China's cultural heritage,which have rich historical value and scientific research value.In recent years,hyperspectral imaging technology can perform more comprehensive recording and analysis of cultural relics with the characteristics of non-destructive and ‘image-spectrum merging'.In the study of painted cultural relics,the use of hyperspectral image to mine implicit information is a research hotspot,which is of great significance for the cultural relics' protection and analysis.Feature extraction is a common method of dimensionality reduction for hyperspectral data.It can mine implicit information from hyperspectral data of painted cultural relics.However,the existing method for feature extraction of implicit information is based on linear transformation,and the hyperspectral image of painted cultural relics has nonlinear characteristics.In order to extract implicit information better,we try to use nonlinear transformation for feature extraction.This paper study the feature extraction method based on neural network and image fusion method.The method we proposed obtained more satisfactory results.The main contributions of this paper are as follows:(1).The multi-scale residuals convolution-deconvolution feature extraction network is designed to deal with the complex pattern information and the absence of implicit information labels in the data of painted cultural relics.In order to adapt to different data characteristics,this paper adds a multi-scale module on the basis of convolution-deconvolution unsupervised network framework.The network makes feature concentration and visualization while extracting feature.Finally,the mined implicit information is fused with the true color image.The implicit information and the spectral information of the band are comprehensively displayed.The experimental results show that the method can mine the implicit information of painted cultural relics more clearly and accurately.(2).In order to improve the quality of feature extraction results of painted pottery,a method based on neural network classification model was proposed.According to the pattern and the background area,which obtained by painted pottery feature extraction,is obvious and the pixel difference is large,the method of automatically designing the label and automatically selectingthe training sample is proposed.The trained neural network classification model is used to further extract the implicit information on the image containing the implicit information.Experimental results show that this method can extract more comprehensive and abundant implicit information.