Optimum Feature and Classifier Selection for Accurate Urban Land Use/Cover Mapping from Very High Resolution Satellite Imagery

Feature selection to reduce redundancies for efficient classification is necessary but usually time consuming and challenging. This paper proposed a comprehensive analysis optimum feature the most classifier accurate urban area mapping. To this end, 136 multiscale textural features alongside panchromatic band were initially extracted from WorldView-2, GeoEye-3, QuickBird satellite images. The wrapper-based filter-based implemented optimally select best ten percent of primary initial set. Then, machine leaning algorithms such as artificial neural network (ANN), support vector (SVM), random forest (RF) classifiers utilized evaluate efficiency these selected classifier. achieved set was validated using two other images WorldView-3 Pleiades. experiments revealed that RF, particle swarm optimization (PSO), neighborhood component (NCA) resulted in methods, respectively. While ANN SVM’s process depended on number input features, RF significantly resistant criterion. Dissimilarity, contrast, correlation played greatest contributing role performance among used study. These trials showed could be reduced 14 137; classifier, can produce an F1-measure about 0.90 different five very high resolution sensors various geographical landscapes. results successfully achieve our goal assisting users by eliminating task optimal thereby increasing land use/cover also computational load feature-engineering phase deep learning approaches.