The Research Of Endmember Extraction From Hyperspectral Image Based On Mathematic Morphology

The development of hyperspectral remote sensing began in the 1980s, as the progress of imaging technology, hyperspectral remote sensing images with their advantage have been widely used in more and more research field. Comparing with the traditional multi-spectrum remote sensing image,hyperspectral image contains hundreds of continuous bands of data .As getting the ground information ,it gets the continuous spectrum information of every pixel in a spectral interval. So that each pixel can extract a continuous smooth spectrum curve,solved the technical problems of"image without spectrum"and"spectrum not imaging".However,because the limit of imaging spectrometer and the complex of the earth surface,a pixel in the image always contains many different terrain types,which formed a mixed pixel,and it's a mistake to divide it to any class. If a pixel contains only one pure features,it is called endmember.It'a urgent problem to extract endmembers from all the pixels in the image.Then,the following technology can be realized:linear unmixing,spectrum matching,image analysis. In recent years,domestic and foreign scholars have made some unmixing model of mixed pixel,such as linear model,geometic optical model,nonlinear model and so on. The advantages of linear mixed model is simple in structure,clear in physical meaning,and effectively accurate to extract the endmembers.Since the birth of mathematic morphology ,it has received more concerned and deep research,and it is widely applied in many fields.It was used in GIS in 1980s,such as feature extraction of remote sensing,edge detection,image segmentation ,etc.But its relatively less used in hyperspectral .This paper is based on the imaging theory of hyperspectral mixed pixel,and introduce the expression of hyperspectral data,the mixed model of pixel,and the classical algorithm of endmember extraction.Then,research endmember extraction on the premise of linear mixed model. This algorithm expand mathematical morphology so that it can be used in the extraction of endmember.This algorithm is different from the traditional algorithm which ignore the space information in the image,it organically combine the spectrum information with space information.It considers the internal space characteristics when using the spectrum information,so that research and improve the traditional algorithm.Theoretically guide by mathematical morphology,expanding the basic definition of 2-D image corrosion and dilation to multi-dimensional remote sensing images.Finally,do experiment by computer simulation,and find that the method indeed has better effect.