Feature selection for urban land cover classification employing genetic algorithm

Abstract

Feature selection has attained substantial research interest in image processing, computer vision, pattern recognition and so on due to tremendous dimensional reduction in image analysis. This research addresses a genetic algorithm based feature selection strategy for urban land cover classification. The principal purpose of this research is to monitor the land cover alterations in satellite imagery for urban planning. The method is based on object based classification by detecting the object area of a given image with the knowledge of visual information of the object from remote sensing images. The classification system is organized through a multilayer perceptron with genetic algorithm (MLPGA). Experimental results explicitly indicate that this MLPGA based hybrid feature selection procedure performs classification with sensitivity 94%, specificity 90% and precision 89%, respectively. This MLPGA centered hybrid feature selection scheme attains better performance than the counterpart methods in terms of classification accuracy.