Assessment of Machine Learning Methods for Urban Types Classification Using Integrated SAR and Optical Images in Nonthaburi, Thailand

Urbanization and expansion in each city of emerging countries have become an essential function Earth’s surface, with the majority people migrating from rural to urban regions. The various category characteristics emphasized great importance understanding creating suitable land evaluations future. overall objective this study is classify zone utilizing building height which estimated using Sentinel-1 synthetic aperture radar (SAR) satellite-based indexes Sentinel-2A. first research estimate SAR Nonthaburi, Thailand. A new indicator, vertical-vertical-horizontal polarization (VVH), can provide a better performance, produced dual-polarization information, vertical-vertical (VV), vertical-horizontal (VH). Then, model was developed indicator VVH reference data. root means square error (RMSE) between 1.413 m. second three classes types, are composed residential buildings, commercial other including vegetation, waterbodies, car parks, so on. Spectral indices such as normalized difference vegetation index (NDVI), water (NDWI), built up (NDBI) extracted Sentinel-2A To three-machine learning classifier, support vector machine (SVM), random forest (RF), k-nearest neighbor (KNN) were developed. classification uses randomly trained data 500 m focus divided into 100 × grid. Different models examined different variables, for example, only spectral indices. used types. Not average grid used, but also minimum, maximum, mean, standard deviation calculated NDVI, NDWI, NDBI, height. There total 16 variables model. Eventually, principal components analysis (PCA) reduce get performance models. SVM showed accuracy than two, RF KNN. accuracies SVM, RF, KNN 0.86, 0.75, 0.76, respectively.