Researches On Hyperspectral Image Classification Algorithms Based On Sparse Representation

With the rapid development of hyperspectral remote sensor technology,hyperspectral image(HSI)is widely used in many fields,including the hyperspectral image classification technology,which has become a hotspot of hyperspectral remote sensor technology.Hyperspectral image has high spatial resolution,each pixel contains hundreds of band spectral information,which is of great help to improve the accuracy of ground targets classification.Meanwhile,how to make full use of spatial information and spectral information to further improve the classification accuracy is also very challenging.Different from traditional hyperspectral image classification methods,the sparse representation classification method uses a over-complete dictionary to sparse representation of signal,then classifies by sparse coefficients,which leads to a good classification performance.Based on the background of the existing hyperspectral image classifications and sparse representation in the latest theory,the paper makes a deep research of the hyperspectral classification method based on sparse representation and has achieved a certain significant result.The paper mainly contains:1.The hyperspectral image classification method based on sparse representation:Firstly using spectral information for sparse coefficient by solving algorithms,and processing a series of sparse coefficient for probability graph,then using spatial information to filter processing,and finally comparing the probability graph to get the classification results.2.The multiple features based on sparse representation method of hyperspectral image classification:combined with kernel transformation of sparse representation to classify the information transform of low dimensional space to high dimension space,which leads to good effects in some categories.At the same time to classify the original information of hyeperspectral image,which also gets good results in some categories.Therefore,through a combination of liner and nonlinear features,to use hyperspectral image spatial information and spectral information at the same time on the classification.This paper sets on contrast experiment of the hyperspectral image classification classic data in Indian Pines、Pavia University and Salinas Scene to verify the effectiveness of the proposed method.Experimental results show that the presented methods have obtained improvement in classification accuracy.