Code for:

CNN-Based Semantic Change Detection in Satellite Imagery, Ananya Gupta, Elisabeth Welburn, Simon Watson & Hujun Yin, ICANN 2019

Post Disaster Mapping With Semantic Change Detection in Satellite Imagery, Ananya Gupta, Elisabeth Welburn, Simon Watson & Hujun Yin, CVPRW 2019

Our work focuses on identifying road networks and buildings in post-disaster scenarios using publicly available satellite imagery and neural networks for segmentation. We use a change detection framework to identify areas impacted by the disaster and use inspiration from graph theory to update road network data available from OpenStreetMap in the aftermath of a disaster.

• Download data of interest from DigitalGlobe OpenDataInitiative

• Download Labels from OpenStreetMap

• Use QGis to extract any polylines marked as roads,trunk, link etc (essentially extract all road polylines) and export as geojson

• Use tools provided by SpaceNet apls to generate raster files from OSM vector data

Segmentation training code based on TernausNet. Added models, updated training scripts:

python aerial_train.py

This script includes the following models and backends:

• 'UNet11': UNet (VGG11),
• 'UNet16': UNet (VGG16),
• 'UNet18': UNet (ResNet18)
• 'UNet34': UNet (ResNet34)
• 'UNet11Upsample' : UNet (VGG11 with linear upsampling),
• 'UNet16Upsample' : UNet (VGG16 with linear upsampling),
• 'UNet18Upsample' : UNet (ResNet18 with linear upsampling),
• 'UNet34Upsample' : UNet (ResNet34 with linear upsampling),
• 'LinkNet18': LinkNet (ResNet18),
• 'LinkNet34': LinkNet (ResNet34)

Generate road, building, combined segmentation masks using trained models

python generate_masks.py

Convert segmentation masks to graph, generates metrics and output files

extract_osm_diff.py

Please consider citing the following if you find this work useful:

@article{gupta2020deep,
title={Deep Learning-based Aerial Image Segmentation with Open Data for Disaster Impact Assessment},
author={Gupta, Ananya and Watson, Simon and Yin, Hujun},
journal={arXiv preprint arXiv:2006.05575},
year={2020}
}

@inproceedings{gupta2019cnn,
title={CNN-Based Semantic Change Detection in Satellite Imagery},
author={Gupta, Ananya and Welburn, Elisabeth and Watson, Simon and Yin, Hujun},
booktitle={International Conference on Artificial Neural Networks},
pages={669--684},
year={2019},
organization={Springer}
}

@inproceedings{gupta2019post,
title={Post Disaster Mapping With Semantic Change Detection in Satellite Imagery},
author={Gupta, Ananya and Welburn, Elisabeth and Watson, Simon and Yin, Hujun},
booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops},
pages={0--0},
year={2019}
}