| Hyperspectral imagery contains abundant spatial information and spectral information integrated with image and spectra.Each pixel in space with continuous spectral feature on hundreds of band which is the important basis to accomplish imagery analysis.Hyperspectral imagery contains geometrical characteristic and spectral categorical attributes are benefit for target detection.Research the imagery analysis method is useful to accomplish the hyperspectral imagery data mining.The hyperspectral imagery analysis method mainly contain: band selection,endmember extraction and classification respectively.Band selection is preprocess procedure to select the higher signal noise ratio(SNR)image without redundant information.Endmember extraction confirm the pure spectra as the basis signal to decompose the mixed pixels with subpixel target detection or classification.Classification algorithm provided the basis for region segmentation or made strategic decision.Classical hyperspectral imagery analysis method lacks of dynamic integration about spatial and spectral information,and the scientific parameter index for the number of bands,endmembers,classes.To solve the problem mentioned above,this paper systematic purposed the hyperspectral imagery analysis method with superpixel segmentation algorithm,promoted the utilization of hyperspectral imagery spatial and spectral information,reduced the calculation and enhanced the global optimization ability.The first chapter provided a brief introduction about hyperspectral remote sensing,with the research of band selection,endmember extraction,classification.The second chapter illuminated the advantage of density peaks clustering algorithm compared with other classical clustering algorithm with domestic and overseas research.Form the third to the fifth chapter,each chapter contained algorithm principle,experimental data,evaluation index,analysis discussion section,corresponding the hyperspectral imagery analysis problem mentioned above.Overall,the paper made the following progress:(1)Hyperspectral imagery with strong correlation and great redundancy.The band selection based on density peaks clustering algorithm with scientific parameter index improved the band selection and classification result,overcome the slowly convergence speed and weakly global optimization ability.(2)The endmember extraction based on density peaks clustering algorithm,following linear spectral mixing model,to evaluated the number of endmembers in hyperspectral imagery,with the distribution feature of mixing pixel in high dimensional space.(3)The clustering of density peaks and superpixel segmentation algorithm contained the spatial and spectral information in hyperspectral imagery,using the superpixel instead of pixel to be the basic clustering unit,which is beneficial for the speed and efficiency of this algorithm.Finally,the sixth chapter summarize the contribution of this paper and point out the next research direction. |
