Linear pixel unmixing is a straightforward and efficient approach to the spectral decomposition of multichannel&hyperspectral remotely sensed scenes.A main drawback to its utilization in operational cases is that the endmember of spectral comˉponents can not be retrieved correctly and automatically.Developing unsupervised methods to automatically abstract endmember is a difficult but significant job.The authors presented an iterative error analysis algorithm to retrieve endmembers and unmixing hyperspectral imagery automatically after obtaining some constraint conditions of selecting endmembers by analyzing error propagation in linear spectral unmixing model

and combined with the property of endmemberwhich is cohesive in spatial.The experimental results show the algorithm is robust by testing various thresholds and initial iterative value.Other experiments for test efficiency and accuracy of the algorithm by employing AVIRIS and PHI hyperspectral data were also done.