Spectral Unmixing And Adhibition For Hyperspectral Image Based On Linear Mixed Model

Starting in the80s, hyperspectral remote sensing technology has began to grow,and later as the imaging technology matures, with its own advantages, Hyperspectralremote sensing images has been widely used in more and more research field.Contrasting with the traditional multispectral remote sensing image, hyperspectralimage obtained the continuous spectrum information of each pixels in a spectralinterval, which solved the " image without spectrum "," spectrum not imaging "technical problems.Due to the resolution of the imaging spectrometer is not very high,which makes the image of each pixels, often contain a variety of different terrain types,which formed mixed pixel,it’s a mistake to divide it to any class in the process ofinformation extraction.The only contains a kind of pure object pixels, known as theendmember.It has become research focus to extract accurately endmembers fromhyperspectral image, then, the following technology can be easy realized:linearunmixing,spectrum matching,image analysis, etc. in recent years, domestic and foreignscholars have made some unmixing model of mixed pixel,such as linear mixed modelwith its simple structure and the advantages of physical meaning have become a hotspot of research.This paper first introduces the characteristics of hyperspectral image dat, thedimension reduction of hyperspectral image, the classic algorithms of extractedendmember, the theory of linear mixed model, etc.In terms of dimension reductionmainly adopted principal component analysis(PCA) and minimum noise separation(MNF), and adopted PPI and N-FINDR to extracted endmember.Then，studied thelinear spectrum separation and MTMF, which linear spectrum separation mainlyadopts the constrained least square method to evaluate abundance. Finally, the maintheoretical of MTMF is mainly introduced, with combining MNF transform, takingadvantage of matched filtering for abundance estimate, and MT checking out andeliminate false positive values. The experimenta results found the error of abundanceestimate using MTMF method is smaller. Meanwhile, the experiment of hyperspectralimage data in ENVI under the MTMF mrthod has good effect. Finally, MTMF methodis applied to the hyperspectral SASI data by flight measurement，which was obtainedby the airborne CASI/SASI hyperspectral remote sensing measurement instruments in JiMuSaEr area, decomposition of mixed pixels for the identified endmembers based onMTMF method, and alteration mapping. The results with the geological features whichGoogle Earth to find to the identification of a mapping field are basic consistent.