Urban building detection using object-based image analysis (OBIA) and machine learning (ML) algorithms

Abstract

The information on building features especially in the urban area is very important to support urban management and development. Nevertheless, the automated and transferable detection of building features is still challenging because of variations of the spatial and spectral characteristics to support urban building classification using remote sensing techniques. Most previous studies utilized high-resolution images to discriminate buildings from other land use in the urban area and indeed it involves a high cost to achieve that purpose. Consequently, this study utilized a medium resolution remote sensing image, Sentinel-2B with a 10-meter spatial resolution to classified the building in Selangor, Malaysia. In order to obtain a good classification accuracy, the suitable segmentation parameters (scale, shape and compactness) and features selection for building detection have been determined. Machine learning (ML) algorithms, namely Support Vector Machine (SVM) and Decision Tree (DT) classifiers have been applied to categorized five different classes which are water, forest, green area, building, and road. The result from these two classifiers was then have been compared and it is obviously showing that the SVM classifier is able to produce 20% better accuracy compared to the DT classifier, with 93% and kappa is 0.92. Thus, by enhancing the classification techniques in OBIA, building extraction accuracy using ML algorithms for medium resolution images can be improved and the expenses can be reduced as well.