Remote Sensing Image Decomposition Of Mixed Pixels, The New Method And Application Research

Under the background of globalization, remote sensing technology has been widely used in agriculture, forestry, geology, geography, hydrology, meteorology, marine, mapping, environmental protection, military reconnaissance and many other areas. Remote sensing technology plays a more and more important role in support of decision-making, to serve the economic construction and social development, to deal with unexpected natural disasters. Remote sensing image provide rich information hitherto unknown to human beings, which the spectral properties and the spatial characteristics of the image are organically combined together. With the technology and method of the development of remote sensing technology, the remote sensing data were transformed from the sensors into available information, which has made considerable progress.But because the remote sensing image limited spatial resolution of sensors and the features of the complex diversity lead to mixed pixels exist in remote sensing image. That result in a series problems of remote sensing image feature extraction and the application of remote sensing technology and becomes quantified hampered. Pixels decomposition of mixed pixels analysis method is the most effective, has made many important achievements and applied to the remote sensing image software. Mixed pixel decomposition relates to pure endmember spectra acquisition and endmember abundance estimation of two key issues, the former was mixed pixel object class information, the proportion of all kinds of endmembers obtained in mixed pixel in the unmixing process. The study includes:1. Based on the analysis of the reasons for the formation of mixed pixels, the preprocess the images of remote sensing is provided from combining with image segmentation algorithm in two watershed transformation and Normalized Cut theory. Segmentation areas are obtained based the spectral information and spatial information of mixed pixel.2. Data field theory is proposed to introduce to the mixed pixels endmember extraction. Through processing of remote sensing image based on combination of the second watershed transformation and Normalized Cut, segmentations are closed. These segmented regions are quantitative analysis of radiant energy of remote sensing image pixels in each zone. The endmembers are extracted based on data field theory. 3. Through the research of the application of support vector machine theory in the pixel unmixing, the advantages of the weighted posterior probability support vector machine theory in pixel unmixing is presented. Considering the difference of each support vector machine classifier, posterior probability pixel is used as weight coefficient of subpixel classification for pixels unmixing.4. Analysis the city land use change and dynamic monitoring of land use change trend based on the1995,2005and2010TM remote sensing data of Linyi city. Using the data field theory and Normalized Cut combined with watershed transformation to extract endmembers. Weighted posterior probability support vector machines is used to decompose the mixed pixels of remote sensing image mixed pixel decomposition and the city land use information extraction is extracted.The innovations in the dissertation as bellow:1. Through the analysis of cause of formation of mixed pixels, data field theory is introduced into the endmember extraction. Combined with the second watershed transformation and Normalized Cut theory to segment the image pretreatment of remote sensing image, forming an endmember extraction model based on the spatial and spectral information of mixed pixel.2. Through the research on the theory of support vector machine applied in the mixed pixel unmixing model, considering the differences between all two types of support vector machine classifier. Posterior probability is used as the two class support vector machine classifier’s weight coefficient. The method is applied to the mixed pixel unmixing model.3. The proposed endmember extraction method based on data field theory which combined Normalized Cut and watershed transformation is used to determine the city land use type end element. The weighted posterior probability support vector machine theory is used to unmix the remote sensing image mixed pixels. The city land use information extraction and dynamic monitoring the trend of land use change are based on sub-pixel.