Research On Hyperspectral Image Classification Algorithms Based On Spectral-Spatial Feature And Dictionary Learning

In recent years,sparse representation and cooperative representation have been successfully introduced into the classification of hyperspectral images,which achieved good classification performance.However,hyperspectral images have the characteristics of high dimension and high correlation between bands.There are also the problems of same spectrum and different classes in the classification process.How to extract the effective features of hyperspectral images and improve the classification accuracy is an arduous task in classification.This paper studies the local structure relationship and feature extraction,and puts forward the following two improved algorithms:1.Aiming at the problem of low classification accuracy caused by the traditional sparse representation failing to make full use of spatial information,weak local constraint ability of dictionary,high correlation,and difficult separation of samples,a hyperspectral image classification method based on hierarchical network and local constraint is proposed in this paper.Firstly,the layered depth network is trained by spatial spectrum information to fully obtain the spectral and spatial information of hyperspectral images.Secondly,the k-nearest neighbor method is used to obtain the local information of the image and generate a dictionary.Finally,spatial spectrum information and local constraint information are combined to extract the features of high-dimensional data.At the same time,the intra class similarity of training samples and test samples is fused to improve the accuracy of classification.Experimental results show that the classification performance of this algorithm is much higher than that of other classification algorithms.2.Aiming at the problem that the traditional collaborative representation can not extract and utilize the features of hyperspectral images well,this paper proposes a spatial spectrum fusion feature and Laplace mapping kernel collaborative representation algorithm.The proposed method is also applies to the classification task of hyperspectral images.Firstly,the depth network is used to fully learn and fuse the spatial spectral features of hyperspectral images.Secondly,the kernel mapping technology is used to map the high-dimensional image to characterize the locality and label information of hyperspectral image.Finally,the local popular information of dictionary atoms is obtained by manifold learning and the constructed Laplace kernel model is integrated into the collaborative representation classification model as a priori information.The experimental results show that the proposed method is more concise and makes full use of the local manifold structure,which can achieve better results in the classification of hyperspectral images.