Unsupervised Orthogonal Subspace Projection Approach to Unmix Hyperspectral Imagery Automatically
非监督正交子空间投影的高光谱混合像元自动分解

Linear pixel unmixing is a straightforward and efficient approach to spectral decomposition of remotely sensed data. In recent years, Orthogonal subspace projection approach has been investigated and used in Linear pixel unmixing widely since it was proposed several years ago. A main drawback to its utilization in operational cases is that the spectral priori knowledge can not be automatically retrieved correctly and completely. To overcome the problem of not knowing the prior endmembers in an image dataset, this paper presents an unsupervised orthogonal subspace projection (UOSP) algorithm to retrieve endmember automatically at each time by searching the maximal pixel vector in an orthogonal imagery. If the pixel satisfied the property of being cohesive in spatial, it would be regarded as an endmember, then was removed the effect of it by orthogonal subspace projection method to get another orthogonal imagery. The experimental result shows that UOSP algorithm is an efficient and precise approach to retrieve endmembers and unmixing hybrid pixel automatically by employing PHI hyperspectral data.