Multi-spectral Remote Sensing Images Classification Method Based on SVC with Optimal Hyper-parameters

Traditional classification methods based on asymptotic theory for multi-spectral remote sensing images need the infinite training samples, which is impossible to be satisfied. And it has massive data information. Support vector classification(SVC) based on small samples overcomes above problems. However, the parameters determining its structure need to be optimized. For that, three optimization algorithms including genetic algorithm, particle swarm optimization and adaptive chaotic culture algorithm, are introduced to obtain optimal hyper-parameters of SVC model for multi-spectral remote sensing images. Experimental results compared with cross-validation method indicate that the computation time for classification by genetic algorithm is least and the generalization of genetic algorithm-based SVC model is best.