Hyperspectral Image Supervised Classification Method And Its Application In Painting Heritage Conservation

With the continuous development of spectral imaging technology,hyperspectral image technology has been successfully applied in many areas.An important feature of hyperspectral image data is "spatial and spectral in one map".When captured the spatial dimensional information,the spectral information of the target object was also collected.Supervised classification of hyperspectral image data is an important aspect of the data analysis,this paper studies the only spectral feature and spectral-spatial feature of the supervised hyperspectral image classification techniques.The main contents are as follows:Firstly,introduced the imaging theory and data characteristics of hyperspectral images,described its important role in many different application areas and also the research background and significance.Secondly,described the extraction methods of hyperspectral image data spectral and spatial feature.In the spectral feature extraction process,principal component analysis,the minimum noise fraction transformation,singular value analyses are mentioned.In the spatial feature extraction process,PCA and convolution neural network are considered.Then it introduced the supervised classification algorithm based on the characteristics of hyperspectral data,mainly described the flow of support vector machine,partial least squares method,artificial neural network,deep belief network.Based on the extracted spectral and spatial feature,the optimization supervised classification model is trained and can improve the classification accuracy.Finally,the feature extraction and supervised classification method are used in painting class heritage protection and evaluation.Include: 1)analysis of flaking disease of mural based on near-infrared hyperspectral images;2)analysis of bone disease of architectural paintings based on visible hyperspectral images;3)fake modern Chinese painting identification based on spectral-spatial feature fusion on hyperspectral image.As the flaking disease of mural and bone disease of architectural paintings,first collect the spectral information in different forms of the disease,then use the supervised classification model to do the regression analysis of spectral information on the disease and disease extent performance,what's more,use the optimal analysis model to predict the hyperspectral images pixel by pixel,getting the distribution map of flaking and bone disease which has an important role for disease analysis and prevention.Finally,this paper presents the near-infrared spectral and spatial feature fusion to compose the classification feature,the support vector machine was used as the classifier to do the identification of fake modern Chinese paintings.