부산대학교 도서관

One of the major topics in photogrammetry is the automated extraction of building from data acquired by airborne sensors. What makes this task challenging is the very heterogeneous appearance and dense distribution of buildings in urban areas. While many dataset has been established, none of them pay attention to developing cities where buildings are not well planned. To complement the development of building extraction algorithms, a dataset of high resolution satellite image is constructed in this paper covering Cau Giay district, Hanoi, Vietnam. The dataset consists of 2100 images of size 1024× 1024 pixels extracted from Google Earth. Shape, size, and construction material differ greatly from building to building, thus make it challenging for state-of-the-art algorithm to accurately extract building location. Some baselines are provided using Convolutional Neural Networks (CNNs). Experimental results show that U-Net model trained with Mean Square Error loss is able to achieve comparable results (OA = 92.04).