A novel optimization parameters of support vector machines model for the land use/cover classification

Nowadays, support vector machines (SVM) are receiving increasing attention in land cover/use classification although one of the major drawbacks of the technique is the kernel function selection and its parameters setting. In this paper, a novel SVM parameters optimization method based on self-adaptive mutation particle swarm optimizer (SAMPSO-SVM) is proposed to improve the generalization performance of the SVM classifier. The SAMPSO algorithm, which is based on the variance of the population's fitness, can break away the local optimum by the operation of self-adaptive mutation. Accordingly, very high classification accuracy will be achieved with the best value of the parameters of SVM, which have been searched using SAMPSO. In order to verify the validity of this SAMPSO-SVM method, a remote sensing land use/cover classification model is constructed using multi-spectral Landsat-5 TM data. In particular, they are organized so as to test the sensitivity of the SAMPSO-SVM model and that of the other reference classifiers used for comparison, i.e. maximum likelihood classifier (MLC), SVM classifier and standard PSO algorithm for SVM parameters optimization model (PSO-SVM). On an average, the SAMPSO SVM model yielded an overall accuracy of 93.59% against 83.92% for maximum likelihood classier and outperformed PSO-SVM classier in terms of accuracy (by about 2%). The obtained results clearly confirm the effectiveness and robustness of the SAMPSO-SVM approach to the remote sensing land use/cover classification.