| Data mining on high-dimensional data is a research hotspot in machine learning.Facing the "curse of dimensionality" problem,various scholars have proposed different solutions,among which sparse subspace clustering algorithm is an effective way to solve this problem.High-dimensional data generally has properties such as sparsity,high-dimensionality,and noise.Hyperspectral remote sensing image data is a typical high-dimensional data.Clustering hyperspectral images is beneficial to realize the division and survey of features,and realize the demand of people to perceive distant features through technology.The spectral data of hyperspectral remote sensing images consists of hundreds of continuous and narrow spectral bands,and there is a strong correlation between the bands.Therefore,in the process of realizing the image feature division,in order to improve the division accuracy of the image,it is necessary to overcome the problems of high-dimensionality and noise in the data.Since the existing SSC algorithm for hyperspectral images uses the spectral information of the image to cluster the images without showing more consideration for the spatial information of the image,the clustering effect of the image is not ideal.In order to improve the performance of sparse subspace clustering of hyperspectral images,it is necessary to fully fuse the spectral information and spatial information of the images.This article mainly discusses how to improve the accuracy of sparse subspace clustering algorithm for hyperspectral images from two perspectives: the affinity matrix reconstruction based on information fusion and the weighted sparse model.The main work of this paper is summarized as follows:(1)In order to fully consider the spatial structure characteristics of hyperspectral remote sensing images and improve the clustering accuracy,this paper proposes sparse subspace clustering algorithms GKD and PGKC based on reconstructed affinity matrix.The two algorithms merge the spectral and spatial information of the image,and solve the problem that the SSC algorithm show less consideration for the spatial information of the image,thereby increasing the algorithm of clustering accuracy in hyperspectral images.Spectral clustering is an indispensable link in the process of sparse subspace clustering,and the nature of the affinity matrix of the input spectral clustering directly affects the clustering performance of the algorithm.Therefore,this paper proposes two affinity matrix construction algorithms.Firstly,Principal Component Analysis(PCA)is used to preprocess the pixel data to remove redundant bands.Secondly,the Gaussian similarity of spatial features between pixels is introduced to reconstruct the affinity matrix,thereby obtain an affinity matrix rich in spatial and spectral features.The experimental results show that the GKD algorithm and PGKC algorithm proposed in this paper are better for the two classic hyperspectral datasets can obtain accurate and reliable clustering results.(2)A weighted block sparse subspace clustering algorithm(EBSSC)based on information entropy is proposed.The introduction of block diagonal constraint and information entropy weight can obtain the prior probability that two pixels belong to the same category before the simulation experiment,so that the solution solved by the positive intervention model tends to the optimal approximate solution of the block diagonal structure,making the model obtain the performance against noise and outliers.Such that the model performance is obtained against noise and outliers,thereby enhancing performance of model clustering,to get a better accuracy of the ground feature division.The total clustering accuracy of the EBSSC algorithm on the three classic data sets reached 78.62%,95.27%,and 93.57% respectively.The experimental results show that the clustering performance and clustering effect of the EBSSC algorithm are better than those of the existing classic popular subspace clustering algorithms.The spatial clustering algorithm can effectively solve the problems that conventional subspace clustering algorithm which processing lower Hyperspectral image classification accuracy. |
