Building Footprint Extraction using Deep Learning

The automatic generation of building footprints from satellite images presents a considerable challenge due to the complexity of building shapes. In this project, we have firstly proposed improved generative adversarial networks (GANs) for the automatic generation of building footprints from satellite images. Then, a novel gated graph convolutional network with deep structured feature embedding is proposed to improve the accuracy of the semantic segmentation.

Gated Graph Convolutional Neural Network with Deep Structured Feature Embedding

The latest development in deep convolutional neural networks (DCNNs) has made accurate pixel-level classification tasks possible. Yet one central issue remains: the precise delineation of boundaries. Deep architectures generally fail to produce fine-grained segmentation with accurate boundaries due to their progressive down-sampling. Hence, we introduce a generic framework to overcome the issue, integrating the graph convolutional network (GCN) and deep structured feature embedding (DSFE) into an end-to- end workflow. Furthermore, instead of using a classic graph convolutional neural network, we propose a gated graph convolutional network, which enables the refinement of weak and coarse semantic predictions to generate sharp borders and fine-grained pixel-level classification. Taking the semantic segmentation of building footprints as a practical example, we compared different feature embed- ding architectures and graph neural networks. Our proposed framework with the new GCN architecture outperforms state-of-the-art approaches. Although our main task in this work is building footprint extraction, the proposed method can be generally applied to other binary or multi-label segmentation tasks.

Improved Generative Adversarial Networks

In this work, we have proposed improved generative adversarial networks (GANs) for the automatic generation of building footprints from satellite images. We used a conditional GAN with a cost function derived from the Wasserstein distance and added a gradient penalty term. The achieved re- sults indicated that the proposed method can significantly improve the quality of building footprint generation compared to conditional generative adversarial networks, the U-Net, and other networks. In addition, our method nearly removes all hyperparameters tuning.

Publictions

 [1] Shi Y., Li Q., Zhu X., (2020): Semantic segmentation through a gated graph convolutional neural network with deep structured feature embedding, ISPRS Journal of Photogrammetry and Remote Sensing, 159, 2020, 184-197

 [2] Shi Y., Li Q., Zhu X., (2019): Building Footprint Generation using Improved Generative Adversarial Networks, IEEE Geoscience and Remote Sensing Letters, 16(4), 2019, 603-607.