A HYBRIDIZATION OF AN IMPROVED PARTICLE SWARM OPTIMIZATION AND FUZZY K-MEANS ALGORITHM FOR HYPERSPECTRAL IMAGE CLASSIFICATION
- School of Geomatics and Urban Spatial Informatics, Beijing University of Civil Engineering and Architecture, Beijing, China
Keywords: Hyperspectral Remote Sensing, Land Cover Classification, Fuzzy K-Means, Particle Swarm Optimization, Wetlands
Abstract. A particle swarm optimization (PSO) algorithm has been widely used in the field of remote sensing image classification. We proposed the IPSO-FKM algorithm, which use the improved PSO (IPSO) algorithm to optimize the initial parameters of the Fuzzy K-Means (FKM) clustering algorithm. We combine the crossover operator of genetic algorithms with PSO, and introduce the fuzzy membership degree of fuzzy mathematics into K-means clustering algorithm. Then we use the IPSO-FKM algorithm to optimize the classification results of the Hyperion remotely sensed images, and use FKM, IPSO, and IPSO-FKM to extract the land cover information on the wetlands in Dongting Lakes, China. The experimental results have been validated by the classification results of MLC and the field investigation data. The validation results have been evaluated from three perspectives: the overall classification accuracy and the Kappa coefficient from the pixel perspective, the intra-cluster distance and the inter-cluster distance from the feature perspective, and the partition coefficient and partition entropy from the information perspective. According to the comparison of IPSO and FKM algorithms，the IPSO-FKM algorithm has a better performance than the others in all three respects. Additionally, in terms of the fitness convergence, the IPSO-FKM algorithm has a better searching velocity and better convergence to lower the quantization errors compared with the other two algorithms.