In the rapidly evolving field of remote sensing, the integration of diverse data sources is crucial for enhancing the accuracy and reliability of geospatial information. This paper introduces a novel approach to augment remote sensing capabilities by smartly hybridizing exogenous data sources, thereby improving the overall quality and utility of geospatial analysis. Our methodology leverages advanced Deep Learning techniques to efficiently combine heterogeneous data, such as satellite imagery, with supplementary information from external sources, including social media, news articles, and open data repositories. The application of our smart hybridization approach to natural disaster management, particularly in the context of the Turkey and Syria earthquakes from February 2023, holds immense potential for improving the effectiveness and efficiency of response efforts.
Visualizing social activity density around a location involves creating graphical representations that depict the concentration of various social activities, such as injuries, gatherings, or events, in the vicinity of a specific place. These visualizations often use color-coding or heatmaps to highlight areas of high and low activity. Such tools are valuable for understanding patterns, trends, and potential hotspots in a given area, aiding in decision-making and resource allocation for public safety or event planning.
Comparing pre and post-event impacts through remote sensing images allows for a clear visual assessment of changes, such as natural disasters or urban development. This technique helps monitor environmental alterations, aiding in informed decision-making for disaster response and land management.
Visualizing earthquake-related social media activity over time, revealing trends, reactions, and information dissemination during seismic events, aiding in disaster response and public awareness.
Mapping tweets during a natural disaster offers real-time data, showing affected areas and sentiments. It aids emergency response and public awareness efforts, improving disaster management..
Information visualization employs a graph structure, often a knowledge graph, to analyze and aid decision-making regarding natural disasters. This approach offers a visually intuitive representation of complex data, enhancing our comprehension of disaster-related information. By mapping interconnections and relationships, it enables more informed and efficient decision-making for disaster preparedness, response, and mitigation. It provides a powerful tool for authorities and stakeholders to assess risks and plan effectively, ultimately reducing the impact of natural disasters.
Utilize a knowledge graph to extract event-specific data, identify relevant keywords, and generate a concise summary for enhanced information retrieval and understanding.
Develop a question-answering chatbot utilizing a knowledge graph for event-specific keyword inquiries. The bot parses event data, maps relationships, and provides concise, accurate responses. Enhance event attendees' information retrieval experience.
Information retrieval involves searching for data based on keywords, employing cosine similarity to measure the textual similarity between the query and documents, ultimately helping locate relevant information.
Image segmentation employs text encoding and zero-shot learning, allowing a user to provide a text prompt or image for encoding. This encoded information is then used to generate segmentation for a retrieved image, seamlessly combining text-based instructions and visual data for accurate segmentation results.
@article{Li2022CLIPEventCT,
title={CLIP-Event: Connecting Text and Images with Event Structures},
author={M. Li and R. Xu and S. Wang and L. Zhou and X. Lin and C. Zhu and M. Zeng and H. Ji and S. Chang},
journal={IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2022},
}
@article{fan_disaster_2021,
title = {Disaster {City} {Digital} {Twin}: {A} vision for integrating artificial and human intelligence for disaster management},
journal = {International Journal of Information Management},
author = {Fan, C. and Zhang, C. and Yahja, A. and Mostafavi, A.},
year = {2021},
}
@inproceedings{Zhou2023MultimodalET,
title={Multimodal Event Transformer for Image-guided Story Ending Generation},
author={Zhou, Y. and Long, G.},
booktitle={Association for Computational Linguistics},
year={2023}
}
@misc{nominatim,
author = {S. Hoffmann},
title = {Nominatim : Open-source geocoding with OpenStreetMap data},
year = {2020},
note = {}
}
@INPROCEEDINGS{ai4geo,
author={Brunet, P.M. and Baillarin, S. and Lassalle, P. and Weissgerber, F. and Vallet, B. and Triquet, C. and Foulon, G. and Romeyer, G. and Souille, G. and Gabet, L. and Ferrero, C. and Huynh, T.-L. and Lavergne, E.},
booktitle={2022 IEEE International Geoscience and Remote Sensing Symposium},
title={AI4GEO: A Path From 3D Model to Digital Twin},
year={2022},
}
@inproceedings{ning-etal-2022-meta,
title = "A Meta-framework for Spatiotemporal Quantity Extraction from Text",
author = "Ning, Q. and Zhou, B. and Wu, H. and Peng, H. and Fan, C. and Gardner, M.",
booktitle = "Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics",
year = "2022",
}
@article{ochoa_machine_2021,
title={A Machine learning approach for rapid disaster response based on multi-modal data. The case of housing \& shelter needs},
author={K. Saldana Ochoa and T. Comes},
year={2021},}