The Research On The Classification Method For Hyperspectral Remote Sensing Images

Hyperspectral image(HSI)is the spectral image whose resolution reaches the nanometer scale,and it can detect and identify substances which could not be predicted or distinguished by wide-band remote sensing.It also has important applications in land utilization,resource investigation,natural disasters,global environment,interstellar exploration,etc.The classification of HSI is defined as extracting the spatial characteristics of image by specific classification algorithms.However,higher classification requirements for HIS have been proposed with the increase of spectral dimensions.To enhance classification accuracy of HIS,the improvement of classification algorithms has become an important topic in the research field of remote sensing.In this research,graph-based label propagation algorithm is proposed based on the full study of existing classification methods,and the content of this method includes the graph construction and the label propagation algorithm.For the construction graph,sparse representation model is established by making full use of spatial and spectral information of the image,and the model is also use as measure of action correlation.The graph which can reflect the relation between pixels is constructed after modifying the correlation by residuals.For the label propagation algorithm,considering two objects.(1)two pixels with large similarity have small difference in membership vector.(2)the membership of the sample pixels is consistent with its known class.The label propagation model is weighted summed by two objects mentioned above,and then the membership degree vector of each pixel is solved.In this thesis,the system of linear equations is solved iteratively to improve calculation efficiency.To verify the reliability of classification algorithm,the remote sensing data from the Indian_pines,the PaviaU in Italy,and the Salinas Valley in California,USA are simulated,and the results are compared with the existing classification algorithms(sparse representation algorithms,support vector machine algorithms)and algorithms of this paper.The conclusions indicated classification accuracy and Kappa index of HIS has been significantly increased.