Support vector machine approach to identifying buildings using multi-temporal ALOS/PALSAR data

Abstract

Accurate detection of built-up areas is valuable for quantifying the level of urbanization and monitoring the decreasing amount of agricultural land. In this study, three-temporal, dual-polarization Advanced Land Observing Satellite (ALOS)/Phased Array type L-band Synthetic Aperture Radar (PALSAR) images were integrated to map buildings in the Yangtze River Delta of East China, where land has been intensively used. The results show that the support vector machine (SVM) classifier performs well in identifying buildings, with an accuracy of 90% in urban areas and 95% in rural areas, even with only a small number of training samples. Buildings in urban areas are more likely to be underestimated (commission error of 15%) than those in a rural environment. Visual inspection and quantitative analysis confirmed that the Local Sigma Filter considerably reduced random speckle noise in the PALSAR imagery. Thus, the filter is suitable for enhancing feature extraction of future multi-polarization and multi-temporal SAR imagery. Overall, the buildings identification approach proposed in this study could serve as a valuable tool for operational monitoring of rural land use change and urban sprawl.