Research On Hyperspectral Image Classification Method Of Joint Multi-Characteristic Similarity Metric Relations

Hyperspectral image(HSI)contains rich ground objects information,and has been widely used in environmental detection,target recognition and other fields.In image classification process,the traditional classification methods simply exploit spectral or spatial information of HSI,but they ignore various information of image data so that they has misclassification of pixels of HSI.Therefore,how to use various information of image data to classify ground objects and how to improve classification accuracy of HSI is urgent problems at present.Based on multiple information fusion and sparse representation,this paper mainly studies from multiple information fusion for image classification method of HSI.The main research contents are as follows:This paper briefly introduces spatial feature extraction methods.Several traditional classification algorithms are introduced from three different aspects of spectral information,representation model and classifiers combination.And it also introduces several HSI classification evaluation indicators and two HSI data sets in this paper.A new classification algorithm called combination of sparse characteristic and neighborhood similarity metric(SNMC)is proposed,it exploits traditional sparse representation model to calculate sparse coefficient,so test samples can be represented by characteristics of image data,so that it can construct neighborhood similarity of data set.Finally,sparse relationship and neighborhood relationship of data set can be combined,sparse-neighborhood similarity value of test samples can be calculated,and the class of test samples are determined according to sparse-neighbor similarity.Experimental results on the Indian Pines and Pavia University data sets show that the SNMC algorithm has higher classification accuracy than other traditional classification algorithms,and it can significantly improve the classification performance of HSI.A new image classification algorithm called joint multiple feature similarity metric(MCSM)is proposed.The algorithm exploits rich information of HSI,firstly,it needs to extract the spatial features of all the pixels of HSI.Then,spatial similarity can be construct to exploit the rich spatial information of HSI,so spatial similarity probability value can be calculated.Sparse similarity relationship and collaborative similarity relationship are construct by exploiting sparse characteristics and collaborative characteristics of HSI.Finally,spatial-sparse-collaborative similarity is constructed,the land cover types can be obtained with the similarity value.From the experimental results of Indian Pines data set and Pavia University data set,it can be concluded that MCSM algorithm has higher classification accuracy and can significantly improve the classification performance.