Urban Change Monitoring in Developing Countries Based on Deep Learning Technique by Utilizing Time Series Imageries of the SAR and Optical Satellites

Abstract

Frequent monitoring of the urban changes is important for urban and economic development planners. Satellite imageries provide chance for mapping the rapid urban expansions and can be effectively used for monitoring the urban changes. This study highlights the automatic extraction methodology of generating land cover maps from single-polarization SAR imageries of TerraSAR-X (HH) of the developing regions in Southeast Asia such as Indonesia and the Philippines by utilizing deep learning algorithm. As single-polarization SAR imageries have difficulty of classification of water areas, the water areas are extracted from optical satellite imagery of Landsat and Sentinel-2. After generating land cover maps, urban areas are extracted. By analyzing time series of satellite imageries in the proposed methodology revealed that there are fast developments of the urban areas in the developing countries. The results were validated by the existing facts of development plans for each country. This study demonstrated about the value of combining SAR and optical satellite imageries for urban change monitoring by utilizing deep learning technique.