Study Of Semi-supervised Hyperspectral Image Classification Method Based On Spectral-Spatial Information

Hyperspectral image can provide very abundant spectral information which can reflect the physical nature of the different materials.This spectral information can be utilized to distinguish objects.Traditional classification methods are facing many problems and challenges in the classification of hyperspectral images,such as classification with a few samples and improve the execution efficiency of the algorithm,etc.This paper makes full use of the spectral information and spatial information of hyperspectral images to study two semi-supervised hyperspectral classification methods based on the research status of hyperspectral image classification methods.The proposed method was verified using multiple hyperspectral data sets.1.In order to overcome the influence of a few samples on the classification accuracy of hyperspectral images,a new hyperspectral image classification method based on multi-resolution segmentation and edge-preserving filtering is proposed.First,the hyperspectral image was segmented to homogenous regions with multi-resolution segmentation method.The unlabeled pixels were selected randomly to assign the class labels.It can effective to overcome the problem that the classification accuracy of hyperspectral image decreases significantly.Then,use support vector machine to classify hyperspectral images to obtain the initial classification result.Finally,edge-preserving filtering is performed on the classification result to remove the "noise,and correct the classification result.The method makes full use of the spatial correlation between pixels.The proposed method is compared with multiple widely used hyperspectral image classification methods in the experimental.Experimental results show that hyperspectral image classification method based on multi-resolution segmentation and edge-preserving filtering can make full use of the spatial context information.It can increase the number of training samples effectively and improve the classification accuracy of hyperspectral images.2.Traditional label propagation method is a method of local propagation,which ignores the spatial information of hyperspectral images and has a poor classification effect with a few samples.In order to resolve this problems,a new hyperspectral image classification method based on superpixel and label propagation was proposed.First,the hyperspectral image was divided into a large number of superpixels.Increasing the number of training samples by the nature of the superpixels that pixels in the same superpixels spectral similarity and spatially adjacent.Then,labels were propagation from labeled samples to unlabeled samples use the label propagation algorithm.The result of label propagation as the classification result of hyperspectral image.Before the label propagation,a sparse graph is constructed by taking into account the spectral information and spatial information of the hyperspectral image.Each vertex in the sparse graph represents a sample in the graph.The weights of edges between the samples are obtained by combining the spectral information and spatial information.The proposed method breaks the local framework of traditional label propagation by propagation label to all unlabeled samples.In the experiment,the proposed method is compared with multiple hyperspectral image classification methods.Experimental results show that the running time of this method is not affected by the number of training samples.And the classification accuracy has exaltation greatly with a few samples.