Multiple Endmember Spectral Mixture Analysis Based On Medium Spatial Resolution Satellite Imagery

Medium-resolution remote sensing images(TM/ETM+) are widely used in urban researches, but there exist a lot of mixed pixels. Because of strong spatial heterogeneity of urban areas, the mixed pixels are even more in Medium-Resolution remote sensing images of urban areas. Traditional image processing methods of remote sensing (supervised and unsupervised classification), which are based on the pixel level, meet difficulties for mixed pixels. To improve the accuracy of Medium-Resolution image in urban studies, we must solve the problem of decomposition for the mixed pixel.Taking Changsha, Hunan as an example, this study discussed vegetation-impervious surface-soil (V-I-S) model for the problem of mixed pixels in medium resolution remote sensing images. Using pixel multiple endmember spectral mixture analysis (MESMA) to the urban area of Changsha, we can get the output of fractional maps. The study also established a new MESMA method—the area-based MESMA, and study the advantages and disadvantages of the two MESMA methods in the classification of urban land cover. The main works are as follows:(1) Based on the supervised classification and pixel-MESMA, this study developed a new MESMA method, the area-based MESMA. This method was realized in the environment of ENVI 4.1 and IDL 6.1. The results of this method are fractional maps.(2) In the support of ENVI 4.1 and IDL 6.1, the research used the existed MESMA to the study area. The results of land cover classification abundance images, including the vegetation-,the soil-,the low reflection-and the high reflection-abundance images.(3) Selecting optimum endmembers for the study area, based on the V-I-S model and the actual situation. Some transformations were carried out to optimize the selection of endmembers. First, the image dimensionality was reduced through MNF. Then, PPI transformation was used to narrow the range of endmember. The next step, the pure pixels were selected as endmembers by interactive way in the transformed image. In the end we selected four kinds of endmember, including vegetation, soil, low reflectance object and high reflectance object. (4) Using qualitative and quantitative analysis methods to validate the precision of the results of pixel-based MESMA and area-based MESMA, and the quantitative analysis methods include RMSE and Confusion Matrix. This paper discussed the advantages and disadvantages of the two methods in the urban studies. We also compared with the classification results of supervised and fixed-endmember linear spectral mixing analysis (LSMA), and the compilation proved that MESMA is a better method for the decomposition of mixed pixel.(5) At last, pointing out the advantages and limitations of the area-based MESMA in the urban studies and the direction of further research.