Hyperspectral Remote Sensing Image Classification Based On Joint Sparse Representation

Hyperspectral remote sensing is an emerging remote sensing technology.Information of the target objects can be obtained by hundreds or even thousands of band spectral response between visible light and infrared light.With the advantages of the high spatial resolution and high spectral resolution,hyperspectral remote sensing is playing an important role in the fields of target recognition,precision agriculture,hydrological monitoring,city planning,etc.Hyperspectral remote sensing image represents the categories of ground objects accurately as it has abundant spectral information and spatial information.At the same time,it also has a lot of problems.For example,due to the high dimension of hyperspectral image,it is more difficult to process the hyperspectral image,due to large amount of information in hyperspectral image,it is more difficult to identify the small sample objects,the uncertainty factor is introduced in the analysis of hyperspectral data,etc.In order to improve the classification accuracy of hyperspectral remote sensing image,in this thesis,two kinds of highspectral remote sensing image classification algorithms based on joint sparse representation are presented in combination with spatial and spectral information of highspectral remote sensing images : 1.Classification algorithm based on spectral clustering and joint sparse representation.First,pixels and its neighboring pixels are classified into two categories by spectral clustering.And then joint sparse representation model is used to determine the specific categories of one of the categories selected according to the rules,and this category is regarded as the category of the pixel to be tested.Finally,the classification algorithm is corrected by using the spatial information to correlate the categories of neighboring pixels.2.Classification algorithm based on neighborhood similarity and joint sparse representation.The similarity threshold is set according to the correlation degree between the pixels to be tested and pixels in the neighborhood,pixels which have high similarity to the pixels to be tested are represented by joint sparse representation.Then the neighboring voting method is used to correct the classification algorithm,and the categories of the pixels to be tested are determined.In the sparse representation algorithm,the linear combination of the training samples are used to represent the test samples,then the test samples are represented by sparse coefficients.In this way,the image compression is achieved,which greatly simplifies the subsequent processing of the image.At the same time,the sparse representation algorithm does not need to set the statistical characteristics of the test samples,and different constraints are used in the training samples to satisfy different test requirements.Due to these advantages in dealing with high-dimensional data,the sparse representation algorithm is successfully applied to the classification of hyperspectral remote sensing image.Two classification experiments of hyperspectral remote sensing images are used to illustrate the effectiveness of the hyperspectral remote sensing image classification algorithms proposed in the paper.