The Research Of Multiple Endmember Spectral Mixture Analysis Method

TM/ETM+Medium-resolution remote sensing images are widely used in urban land use/land cover, due to the complexity and spatial heterogeneity of urban surface, the issues of mixed pixels are particularly prominent in the medium-resolution remote sensing images of urban areas,which does not only influence the accuracy of image classification, but also greatly hinder the development of quantitative remote sensing of land surface. Therefore, mixed pixels unmixing of remote sensing images and quantitative extraction of land coverage information have attracted interest in recent years.In order to solve the problem of mixed pixels unmixing, the researchers have done a lot of research, multiple endmember spectral mixture analysis is an effective method to solution the problem of mixed pixels unmixing. In this paper, using the Landsta TM medium-resolution remote sensing image of xi’an city, based on Vegetation-Impervious Surface-Soil (V-I-S) Model,investigate the method of multiple endmember spectral mixture analysis.The main conclusions are as follows:1、 Discussion the endmembers selected of mixed pixels unmixing building a candidate endmember spectral library, based on VIS model, reference to the study area land cover actual situation, adding the water-endmember, determined the study area land cover types are water vegetation, soil, and Impervious Surface, combined the scatterplot endmember selection method and Pixel Purity index (PPI) endmember selection method selecte13pure pixel groups, building the candidate endmember spectral library.13candidate endmembers contains two groups of water endmember, two groups of vegetation endmember, three groups of soil endmember and six groups of Impervious Surface endmember.2、To use the enumeration of multiple endmember spectral mixture analysis method, the method of the correlation coefficient of multiple endmember spectral mixture analysis, the method of spectral matching of multiple endmember spectral mixture analysis and linear spectral mixture analysis method, unmixing the mixed pixel of study area, to get the four methods of unmixing of land cover composition abundance map and root-mean-square error map, qualitative and quantitative accuracy assessment of the unmixing results. Comparative analysis indicated the accuracy of the unmixing based on the enumeration of multiple endmember spectral mixture analysis method and the method of the correlation coefficient of multiple endmember spectral mixture analysis are higher, the accuracy of three kinds of multiple endmember spectral mixture analysis method are clearly higher than the linear spectral mixture analysis method.3、 Comparison of the advantages and disadvantages of the three kinds of multiple endmember spectral mixture analysis method, propose method of normalized multiple endmember spectral mixture analysis to quantitative extraction of land coverage information of vegetation, soil and impervious surface, comparative analysis indicated the accuracy of the unmixing based on the normalized multiple endmember spectral mixture analysis method is higher than that three multiple endmember spectral mixture analysis method. Normalized multiple endmember spectral mixture analysis can solve the mixed pixel problem of the high spectral heterogeneity in urban landscape effectively. Therefore, the proposed methodology can be used to extract urban land surface information effectively, and to analyze the urban ecological environment change.