Research On Hyperspectral Image Classification Method Of Feature Fusion Similarity Measure Relations

The hyperspectral image(HSI)contains abundant feature information,which has the characteristics of high spatial resolution,large amount of data,and strong waveband correlation,it brings more possibilities for the improvement of the fineness of feature classification.However,the traditional hyperspectral classification methods do not make full use of the various relationships contained in the data.Therefore,how to make use of the rich information among hyperspectral image data to improve the classification performance and enhance the classification accuracy of ground objects has become a hot issue in the research of hyperspectral image classification.This thesis starts from the spatial information and spectral information of hyperspectral image data.With the support of common feature extraction methods and hyperspectral classification method theories,researches are carried out from the aspects of feature extraction,feature fusion,and kernel functions to explore hyperspectral images.The impact of multiple information fusion in the image classification method.The main research contents are as follows:A method of image classification based on image reconstruction feature fusion(RSFM)is proposed.The method first extracts the LBP feature value of the pixel,and at the same time extracts the spatial neighborhood block of the point,and then removes the neighborhood based on the image label information.For background points,use the spectral distance to obtain the weights of the pixels in the spatial neighborhood block,calculate the spatial reconstruction feature value of the center pixel,and superimpose and fuse the LBP feature value of the pixel and the spatial reconstruction feature value to obtain the reconstructed feature fusion.K nearest neighbor classification to judge the size of the Euclidean distance algorithm for classification.As can be seen from the experimental results,the experimental results show that compared with other traditional classification methods,the RSFM algorithm has a higher classification accuracy and can significantly improve the classification performance of ground objects.A multi-feature fusion kernel hyperspectral image classification method(MFKM)is proposed.The method first extracts the LBP feature and EMAP feature of the data image,and uses the MLR method to fuse the LBP,EMAP and spectral features to obtain the sample point space The similarity value,and then use the kernel function to project the data to the kernel space to solve the sparse kernel similarity value and the cooperative kernel similarity value of the sample points.Construct the space-sparse-cooperative kernel similarity value,and finally classify it according to the joint similarity value.Experimental results show that the proposed MFKM algorithm can achieve better classification performance compared with the related classification algorithm.