Hyperspectral Remote Sensing Image Classification Based On Joint Sparse Representation

Hyperspectral remote sensing image classification is a hot issue in the field of image research and has a wide range of applications.However,with the rapid development of hyperspectral remote sensing image technology,the spatial and spectral resolution of the acquired image have been greatly improved,it makes possible to use Spatial information.It is very difficult to get a good classification result only using the spectral information.How to make full use of the spatial information of hyperspectral images to improve the accuracy is still an important issue that people keep researching and exploring.This paper studies the hyperspectral image classification algorithm based on independent spectral angle.Spectral angle matching is not affected by the vector mode,only affected by the direction of the vector.The spectral angle is used to classify the hyperspectral image to obtain the initial mark of the whole original image.According to the initial label,construct the neighborhood block involved in hyperspectral image classification in the next step,remove the disturbed pixels in the neighborhood block.Classified all the training samples and calculated the independent spectral angle based on the independent training samples then classified the test sample based on independent spectral angle.Experimental results show that the classification of independent spectral angle can effectively describe the spectral features of objects,so as to obtain the better classification results.In view of the shortcomings of the joint sparse representation classification algorithm in mining and fusing the spatial information and spectral information,this paper proposes an improved joint sparse representation classification algorithm.The specific implementation of this algorithm can be divided into two major steps.The first step is to construct the neighborhood space.Through the research on the traditional joint sparse representation classification algorithm,we found that the neighborhood block of pixel often have large number of disturbed pixels.So we used the initial classification based on the spectral angle of the independent classes to get the object category marks of hyperspectral image.According to the map of marks,neighborhood blocks are constructed and the disturbed pixels in the neighborhood are filtered.The second step is to extracted the global spatial information.Hyperspectral images has rich information and complex background.It is difficult to achieve the desired results with a single feature classification,and the combination of multiplefeatures usually can improve the classification accuracy of hyperspectral images.Extracted the spatial features of hyperspectral image.Each feature representation coefficient is separately introduced into the spatial-spectral joint sparse representation model,calculated the independent sparse representation coefficient.This paper presents a residual fusion algorithm,fully considered the diversity of various features.Experimental results show that compared with other algorithms,the classified accuracy of the proposed algorithm is higher and more stable.In this paper,an improved algorithm is proposed based on the research of joint sparse representation algorithm,which not only considers the spatial neighborhood information but also considers the global spatial information of the image.On the basis of fusing spatial and spectral information,take full account of the contribution of various information to improve the classified accuracy of the joint sparse representation classification algorithm.Finally,the advantages and disadvantages of this algorithm are analyzed in the article,which laid a good foundation for future research.