PSO-GA on Endmember extraction for hyperspectral imagery

The existing particle swarm optimization (PSO) and genetic algorithms (GA) could not solve some discrete-valued problems effectively such as Endmember extraction in hyperspectral imagery. Firstly, the theory of particle swarm optimization was reviewed, and a genetic algorithm based Endmember extraction method was analyzed, which combined with the convex geometry theory. Then, a particle swarm optimization genetic algorithm (PSO-GA) on Endmember extraction for hyperspectral imagery was proposed, which improves the genetic algorithm with the theory of local best structure of particle swarm optimization. Finally, the experiments were carried out by simulative and real hyperspectral image, and the results between the PSO-GA and GA were compared and analyzed. The results of experiments proved the convergence rate of PSO-GA is much faster than GA´s.