SATELLITE IMAGE SEGMENTATION USING DIFFERENTIAL EVOLUTION

Abstract

Satellite images are multi-spectral, high resolution images and they have important applications in many fields like agriculture, climate studies, landscape, geology, fishery etc. One of the main applications of satellite images is land cover classification. Satellite image segmentation is a difficult problem as different regions occupy different land cover types. Large regions are occupied by some land cover types like rivers, vegetation, etc. while other land types like roads, bridges, etc. occupy small regions. A fuzzy clustering technique based on differential evolution has been proposed that employs type 2 fuzzy systems for membership representation. The proposed segmentation method is a clustering based segmentation method that uses Xei-Beni index for fitness evaluation in which Euclidean distance measure is used and for membership function type-2 fuzzy system is employed. The proposed algorithm require number of clusters to be specified beforehand. Validity of proposed method is being shown by comparing the value of cluster validity index, silhouette index with well-known algorithms: Fuzzy c-means, variable length genetic algorithm, Type-1 differential evolution and multi-objective genetic algorithm.