Detecting OpenStreetMap missing buildings by transferring pre-trained deep neural networks
May 12th, 2021 by Hao Li
Recently, a new research paper “Detecting OpenStreetMap missing buildings by transferring pre-trained deep neural networks” (Pisl, J., Li, H., Herfort, B., Lautenbach, S., Zipf, A. 2021) has been accepted at the the 24th AGILE conference 2021. The conference will take place virually on June 8 to 11, 2021.
Accurate and complete geographic data of human settlements are crucial for effective emergency response, humanitarian aid and sustainable development. OpenStreetMap (OSM) can serve as a valuable source of this data. As there are still areas being incompletely mapped in OSM, deep neural networks have been trained to detect such areas from satellite imagery. However, in regions where little or no training data is available, the network training remains problematic.
In this study, we proposed a method of transferring a building detection model, which was previously trained in an area well-mapped in OSM, to remote data-scarce areas. The transferring was achieved via fine-tuning the model on limited training samples from the original training area and the target area. We validated the method by transferring deep neural networks trained in Tanzania to a site in Cameroon with straight distance of over 2600 km, and tested multiple variants of the proposed method. Finally, we successfully detected 1192 missing OSM buildings in Cameroon with the fine-tuned model. The results showed that the proposed method led to a significant improvement in f1-score even with as little as 30 training examples from the testing site. This is a crucial feature of the proposed method as it allows to fine-tune models to regions where OSM data is scarce.
Fig.1: Building detection results of the transferred model in a selected area in Batcham, Cameroon.
The proposed model can detect individual buildings by generating bounding boxes around each of them. However, further volunteer contributions are still expected with regard to mapping detailed footprints as well as validation. Therefore, this work contributes to the efforts towards machine-assisted humanitarian mapping methods. Future works will focus on better facilitating and supporting the OSM mapping community by estimating amounts of missing buildings or prioritizing unmapped areas.
More details will be presented at the AGILE 2021 conference. Stay tuned and hope to see you there!
Related master thesis:
- Pisl, J. (2021). Automatic detection of human settlements in rural Sub-Saharan Africa from satellite imagery with convolutional neural networks and OpenStreetMap. Aalto University.
Previous related work, e.g.:
- Herfort, B., Lautenbach, S., Porto de Albuquerque, J., Anderson, J., Zipf, A. (2021): The evolution of humanitarian mapping within the OpenStreetMap community. Scientific Reports 11, 3037 (2021). DOI: 10.1038/s41598-021-82404-z
- Li, H., B. Herfort, W. Huang, M. Zia, A. Zipf (2020): Exploration of OpenStreetMap missing built-up areas using twitter hierarchical clustering and deep learning in Mozambique. Vol. 166, Pages 41-51 https://doi.org/10.1016/j.isprsjprs.2020.05.007
- Herfort, B., Li, H., Fendrich, S., Lautenbach, S., Zipf, A. (2019): Mapping Human Settlements with Higher Accuracy and Less Volunteer Efforts by Combining Crowdsourcing and Deep Learning. Remote Sensing 11(15), 1799. https://doi.org/10.3390/rs11151799
- Li, H., Herfort, B., Zipf, A. (2019): Estimating OpenStreetMap Missing Built-up Areas using Pre-trained Deep Neural Networks. 22nd AGILE Conference on Geographic Information Science, Limassol, Cyprus.
- Chen, J., Y. Zhou, A. Zipf and H. Fan (2018): Deep Learning from Multiple Crowds: A Case Study of Humanitarian Mapping. IEEE Transactions on Geoscience and Remote Sensing (TGRS). 1-10. https://doi.org/10.1109/TGRS.2018.2868748
- Chen, J & Zipf, A. (2017): DeepVGI: Deep learning with volunteered geographic information. Proceedings of the 26th International Conference on World Wide Web Compagnon. Pages 771-772. https://doi.org/10.1145/3041021.3054250