Algorithm Research on Endmember Extraction Combined With Distribution Statistics

摘要

The spatial resolution of hyperspectral sensor is limited and the surface features are complicated, and each pixel contains more material information, resulting in the existence of a large number of mixed pixels. Therefore, the research on endmember extraction method have been becoming a hotspot in the hyperspectral field. The extraction method which is based on specificity analysis is greatly influenced by abnormal points in the image. This paper mainly combines the pixels' density with distance statistics to analyze different types ofpixels, and then use distribution statistic information to remove the interference of pixels with abnormal properties. On this basis, ATGP, VCA, SGA and NMF are used respectively to conduct the endmember extraction, and then analyze and study the accuracy of the endmember extraction algorithm combining with distribution statistics. Then the validity of this algorithm is verified by simulating hyperspectral images and real hyperspectral images.