Sentinel-2 Satellite Imagery for Urban Land Cover Classification by Optimized Random Forest Classifier

Land cover classification is able to reflect the potential natural and social process in urban development, providing vital information stakeholders. Recent solutions on land are generally addressed by remotely sensed imagery supervised methods. However, a high-performance classifier desirable but challenging due existence of model hyperparameters. Conventional approaches rely manual tuning, which time-consuming far from satisfying. Therefore, this work aims propose systematic method automatically tune hyperparameters Bayesian parameter optimization for random forest classifier. The recently launched Sentinel-2A/B satellites drawn provide remote sensing imageries case study Beijing, China, have best spectral/spatial resolutions among freely available satellites. improved with compared against support vector machine (SVM) (RF) default discriminating five classes including building, tree, road, water, crop field. Comparative experimental results show that optimized RF outperforms conventional SVM terms accuracy, precision, recall. effects band/feature number band usefulness also assessed. It envisaged Sentinel-2 satellite image processing can find wide range applications where high-resolution applicable.