Extracting buildings from aerial scene images is an important task with many applications. However, this task is highly difficult to automate due to extremely large variations of building appearances, and still heavily relies on manual work. To attack this problem, we design a deep convolutional network with a simple structure that integrates activation from multiple layers for pixel-wise prediction, and introduce the signed distance function of building boundaries as the output representation, which has an enhanced representation power. To train the network, we leverage abundant building footprint data from geographic information systems (GIS) to generate large amounts of labeled data. The trained model achieves a superior performance on datasets that are significantly larger and more complex than those used in prior work, demonstrating that the proposed method provides a promising and scalable solution for automating this labor-intensive task.
Establishing up-to-date large scale building maps is essential to understand the urban dynamics, such as estimating population, urban planning, and many other applications. Although many computer vision tasks have been successfully carried out with deep convolutional neural networks, there is a growing need to understand their large scale impact on building mapping with remote sensing imagery. Taking advantage of the scalability of convolutional neural networks (CNNs) and using only few areas with the abundance of building footprints, for the first time we conduct a comparative analysis of four state-of-the-art CNNs for extracting building footprints across the entire continental United States. The four CNN architectures namely: Branch-out CNN, fully convolutional network (FCN), conditional random field as recurrent neural network (CRFasRNN), and SegNet, support semantic pixelwise labeling and focus on capturing textural information at multiscale. We use 1-meter resolution aerial images from National Agriculture Imagery Program as the test-bed, and compare the extraction results across the four methods. In addition, we propose to combine signed-distance labels with SegNet, the preferred CNN architecture identified by our extensive evaluations, to advance building extraction results to instance level. We further demonstrate the usefulness of fusing additional near IR information into the building extraction framework. Large scale experimental evaluations are conducted and reported using metrics that include: Precision, recall rate, intersection over union, and the number of buildings extracted. With the improved CNN model and no requirement of further postprocessing, we have generated building maps for the United States with an average processing time less than one minute for an area of size $\sim \!\!{\text{56}}$ $\text{km}^2$. The quality of extracted buildings and processing time demonstrated that the proposed CNN based framework fits the need of building extraction at scale.