Endmember extraction is one of the key problems for mixel classification of multispectral imagery.Existing algorithms based on convex simplex often find endmembers within the whole convex simplex so that their speeds are slower when more samples are used to obtain endmembers.Since only the vertexes of convex simplex are probably endmembers and they must be located in the boundary of convex simplex

the search space will shrink a lot if finding endmembers is performed only within the boundary points of convex simplex.According to this theory

this paper presents the endmember extraction algorithm based on the boundary of convex simplex.The algorithm includes determination of boundary of convex simplex and fast finding endmembers within the boundary points of convex simplex.Experiments show that the algorithm can find endmembers not only correctly but also faster than existing algorithms.