Extensions Of Orthogonal Projection For Finding Endmembers

Hyperspectral remote sensing data is increasingly wide application in military detecting, environmental protection and public security, etc. For its’double characters. Including in image and spectral information. But because of the complexity of the natural space and ground objects, there are some pixels in the images often contain more material information, became the mixed pixels and increased the difficulty of fata analysis.Endmember represents a relatively pure material spectral characteristics, it plays an important role in analysis of the image composition and its content for quantitative. Therefore, how to extract the endmember in hyperspectral image with very important research significance. Among them, some endmember extraction algorithms which based on orthogonal projection theory and the maximum likelihood estimation theory are hot issues in the study of hyperspectral remote sensing data in recent years. ATGP and MEAC algorithm are the representatives of the two theories. Especially ATGP algorithm,it has been widely used in the analysis of spectral mixing. But there are sonme problems and shortcomings about them.This paper based on linear spectral mixture model, to research and improve the endmember extraction algorithms about that two kinds of theory. The main work includes:Firstly, research the endmember extraction algorithm types, analysis the relationship between orthogonal projection algorithm and maximum likelihood estimate algorithm, such as ATGP and MEAC. Then to compare the performance of different algorithms.Secondly, combining with the characteristics of two kinds of algorithms and compared the similarities and differences between OVP algorithm with OSP algorithm. Then in this paper, put forward a new algorithm by improving ATGP algorithm used OVP algorithm,named UVOP algorithm. By extracting the connection between the ATGP and MEAC,the combining with SMEAC、MEAC algorithm improved ATGP and UOVP, put forward the improved algorithm which named ATGP-M and UOVP-M. Then analysis the data of endembers which found by different algorithms.In the final have the experiment in simulated and real hypersoectral images. Select some related endmember extraction algorithms as the comparison group. Then analysis results to verify the performance of the improved algorihm.Theoretical analysis and experimental results show that the algorithms suggested in this paper have the advantages about both Orthogonal projection theory and maximum likelihood estimation theory. And have stronger practicability.