Development of an Automated System for Building Detection from High-Resolution Satellite Images

Detail information of human settlements is crucial for development issues, such as public health, energy development, and disaster risk management. While the methodologies of settlement mapping in urban areas have been developed well, there are still few studies on settlement mapping in rural areas and villages, which are important for urban development in regard to connectivity of development impacts. For the settlement mapping of rural areas and villages, we developed a system of automated building detection from high-resolution satellite images which are provided from Bing Maps. The algorithm of building detection is based on the Convolutional Neural Network, a well-known deep learning method for image recognition. Because the amount of satellite images is enormous, we implemented the algorithm with a high-performance computing with massive parallel processing on the Data Integration and Analysis System (DIAS) in the University of Tokyo. We demonstrated the building detection using the developed system for Yangon, Myanmar.