Automatic Morphological Endmember Extraction Algorithm Based On Orthogonal Subprojection Space

With the continuous development of remote sensing technology,hyperspectral remote sensing gradually reflects the trend of information quantification.Spectral unmixing is becoming the focus and difficulty of hyperspectral remote sensing.Constrained by external uncertainties such as noise,Spectral unmixing often achieves better results in simulated data than true remote sensing data.It is a hot topic in the current research to propose an algorithm that can achieve good experimental results in real remote sensing data.This article starts with the extraction of the pure pixels from real remote sensing data,and uses Hyperion hyperspectral remote sensing data from the Qinghai Lake Basin of Qinghai Province as real experimental data.The main work includes several aspects:(1)Automated morphological endmember extraction are widely used in hyperspectral mixed pixel decomposition.However,AMEE can only delimit a certain range of candidate endmembers,it can not extract the actual endmembers and needs a certain empirical threshold,which presents a great obstacle to the practical processing of linear spectral hybrid decomposition.This article combines orthogonal subprojection space algorithm with automatic morphology algorithm,and using orthogonal projection divergence take the place of spectral angle to determine pixel purity.Improved algorithm is verified from simulated data and real remote sensing data.The experimental results show that the improved automatic morphological algorithm does not need the support of prior data,and the accuracy of extraction is better than that of some extraction algorithm,especially in real data.(2)The idea of iterative algorithm and pixel-by-pixel calculation is relatively large in the endmember extraction algorithm,but the calculation time is too long and the practicality is poor.Internal removal algorithm of initial simplex extracts the vertex from the space simplex.However,the algorithm is greatly affected by the initial vertex selection,which will result in the missing and repeated selection of the monomorphic vertex.This paper improves the original algorithm and verify the reliability of the improved algorithm through simulated data and real data.The results show that the improved method can effectively enhance the robustness and reliability.