Research On Hyperspectral Image Classification Algorithm Based On Sparse Dictionary And Collaborative Representation

With the continuous improvement of image classification technology,sparse representation and collaborative representation have been successfully applied in hyperspectral image classification.However,hyperspectral images have many problems,such as high data dimension,large redundancy and high correlation of spectral information.How to extract useful information from hyperspectral images and improve the classification accuracy is a difficult task.This paper studies feature extraction and local structure relations,and proposes the following two improved algorithms:Aiming to deal with the problems of insufficient learning of space spectrum features and low classification accuracy of highly similar samples existing in traditional collaborative representation methods,this paper proposes a local discriminant embedded kernel collaborative representation algorithm under space spectrum fusion,and applies it to the classification task of hyperspectral images.Firstly,the space spectrum fusion features of hyperspectral images are obtained by using the space spectrum learning network.Secondly,the low-dimensional linearly indivisible images are mapped to high-dimensional images by kernel mapping to make the high-dimensional images become linearly separable.Finally,the local manifold structure of dictionary atoms is obtained by manifold learning,and it is integrated into the collaborative representation classification model as important prior information.The experimental results show that the proposed method can fully obtain the space spectrum features of hyperspectral images,fully learn the local manifold structure,and improve the classification accuracy when processing high-dimensional data.In view of the low classification accuracy of similar samples due to the weak local constraint ability of dictionary,single feature information and difficult separation of similar samples in traditional sparse representation,this paper proposes a locally constrained sparse dictionary representation algorithm based on differential morphological feature extraction,and applied to the classification task of hyperspectral images.Firstly,the differential morphological feature algorithm is used to increase the differences between similar samples of hyperspectral images.Secondly,the k-nearest neighbor method is used to construct a locally constrained dictionary to increase the locality and correlation between atoms of the dictionary.Finally,class correlation algorithm is used to increase the discriminability of dictionary atoms.The experimental results show that the proposed algorithm can improve the classification accuracy of similar samples,make the image boundary recognition clearer,and eliminate the influence of noise to a certain extent.