Day 4

Detailed paper information

Back to list

Paper title Urban Change Detection from Multi-Temporal Sentinel-1 and Sentinel-2 Data using a Pretrained CNN and Post-Processing
  1. Sebastian Hafner KTH - Royal Institut of Technology Speaker
  2. Yifang Ban KTH Royal Institute of Technology
  3. Dávid Kerekes KTH Royal Institute of Technology
  4. Andrea Nascetti Polytecnic Univeristy of Bari
Form of presentation Poster
  • C1. AI and Data Analytics
    • C1.04 AI4EO applications for Land and Water
Abstract text While more and more people are pulled to cities, uncontrolled urban growth poses pressing threats such as poverty and environmental degradation. In response to these threats, sustainable urban planning will be essential. However, the lack of timely information on the sprawl of settlements is hampering urban sustainability efforts. Earth observation offers great potential to provide the missing information by detecting changes in multi-temporal satellite imagery.

In recent years, the remote sensing community has brought forward several supervised deep learning methods using fully Convolutional Neural Networks (CNNs) to detect changes in multi-temporal satellite imagery. In particular, the vast amount of high resolution (10–30 m) imagery collected by the Sentinel-1 Synthetic Aperture Radar (SAR) and Sentinel-2 MultiSpectral Instrument (MSI) missions have been used extensively for this purpose. For example, Daudt et al. (2018) proposed a Siamese network architecture to detect urban change in bi-temporal Sentinel-2 MSI image pairs. Papadomanolaki et al. (2021) incorporated fully convolutional Long Short-Term Memory (LSTM) blocks into a CNN architecture to effectively leverage time series of Sentinel-2 MSI images. Hafner et al. (2021b) demonstrated the potential of data fusion with a dual stream network for urban change detection from Sentinel-1 SAR and Sentinel-2 MSI data.

Although these urban change detection methods achieved promising results on small datasets, label scarcity hampers their usefulness for urban change detection at a global scale considerably. In contrast to change labels, building footprint data and urban maps are readily available for many cities. Several recent efforts leveraged open urban data to train CNNs on Sentinel-2 MSI data (Qiu et al., 2020; Corbane et al., 2020) and the fusion of Sentinel-1 SAR and Sentinel-2 MSI data (Hafner et al., 2021a). In our previous work, we developed an unsupervised domain adaptation approach that leverages the fusion of Sentinel-1 SAR and Sentinel-2 MSI data to train a globally applicable CNN for built-up area mapping.

In this study, we propose a post-processing method to detect changes in time series of CNN segmentation outputs to take advantage of the outlined recent advances in CNN-based urban mapping. Specifically, a step function is employed at a 3x3 pixel neighborhood for break point detection in time series of CNN segmentation outputs. The magnitude of output probability change between the segmented time series parts is used to determine whether change occurred for a given pixel. We also replaced the monthly Planet mosaics of the SpaceNet7 dataset with Sentinel-1 SAR and Sentinel-2 MSI images (Van Etten et al., 2021), and used this new dataset to demonstrate the effectiveness of our urban change detection method. Preliminary results on the rapidly urbanizing SpaceNet7 sites indicate good urban change detection performance by our method (F1 score 0.490). Particularly compared to post-classification comparison using bi-temporal data, the proposed method achieved improved performance. Moreover, the timestamps of detected changes were extracted for change dating. Qualitative results show good agreement with the SpaceNet7 ground truth for change dating. Our future research will focus on developing end-to-end solutions using semi-supervised deep learning.

The research is part of the project ’Sentinel4Urban: Multitemporal Sentinel-1 SAR and Sentinel-2 MSI Data for Global Urban Services’ funded by the Swedish National Space Agency, and the project ’EO4SmartCities’ within the ESA and Chinese Ministry of Science and Technology’s Dragon 4 Program.

Corbane, C., Syrris, V., Sabo, F., Politis, P., Melchiorri, M., Pesaresi, M., Soille, P., Kemper, T., 2020. Convolutional Neural Networks for Global Human Settlements Mapping from Sentinel-2 Satellite Imagery.
Daudt, R. C., Le Saux, B., Boulch, A., 2018. Fully convolutional siamese networks for change detection. 2018 25th IEEE International Conference on Image Processing (ICIP), IEEE, 4063–4067.
Hafner, S., Ban, Y., Nascetti, A., 2021a. Exploring the fusion of sentinel-1 sar and sentinel-2 msi data for built-up area mapping using deep learning. 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS, IEEE, 4720–4723.
Hafner, S., Nascetti, A., Azizpour, H., Ban, Y., 2021b. Sentinel-1 and Sentinel-2 Data Fusion for Urban Change Detection using a Dual Stream U-Net. IEEE Geoscience and Remote Sensing Letters.
Papadomanolaki, M., Vakalopoulou, M., Karantzalos, K., 2021. A Deep Multitask Learning Framework Coupling Semantic Segmentation and Fully Convolutional LSTM Networks for Urban Change Detection. IEEE Transactions on Geoscience and Remote Sensing.
Qiu, C., Schmitt, M., Geiß, C., Chen, T.-H. K., Zhu, X. X., 2020.A framework for large-scale mapping of human settlement extent from Sentinel-2 images via fully convolutional neural networks. Isprs Journal of Photogrammetry and Remote Sensing,163, 152–170.
Van Etten, A., Hogan, D., Manso, J. M., Shermeyer, J., Weir, N., Lewis, R., 2021. The multi-temporal urban development spacenet dataset. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 6398–6407.