🛰️ BiTemporal-StreetView-Damage

Hyperlocal disaster damage assessment using bi-temporal street-view imagery and pre-trained vision models.

---

📌 Introduction

This repository presents a novel framework for bi-temporal street-view image analysis, aimed at advancing hyperlocal disaster damage assessment. We integrate pre- and post-disaster imagery using pre-trained vision and vision-language models to classify and localize disaster impact more accurately.

🔍 Key Contributions

• ✅ Dual-channel model for fusing pre- and post-disaster street-view images.
• 📸 2,249 labeled street-view image pairs, annotated with fine-grained disaster impact.
• 📈 Performance: Accuracy improved from 66.14% (post-only) to 77.11% (bi-temporal).
• 🔥 Grad-CAM visualization confirms the added value of pre-disaster imagery for model focus.
• 🏙️ Enables rapid and fine-grained damage mapping, supporting climate-resilient urban planning.

Figure: Dual-channel architecture for bi-temporal disaster damage assessment.

---

📂 Dataset

You can access the bi-temporal street-view disaster dataset from the following DOI:

📁 Yang, Yifan (2025).
Perceiving Multidimensional Disaster Damages from Street–View Images Using Visual–Language Models.
figshare. Dataset. https://doi.org/10.6084/m9.figshare.28801208.v2

The dataset includes:

• Pre- and post-disaster images
• Location and damage type annotations
• Severity scores (minor, moderate, severe)
• Sample image regions from Horseshoe Beach, Florida, after Hurricane Milton

---

🧠 Paper Reference

📚 Citation

If you use this repository, please cite both the CEUS article and the arXiv preprint.

---

📖 APA Citation (click to expand)

Yang, Y., Zou, L., Zhou, B., Li, D., Lin, B., Abedin, J., & Yang, M. (2025). Hyperlocal disaster damage assessment using bi-temporal street-view imagery and pre-trained vision models. Computers, Environment and Urban Systems, 121, 102335. https://doi.org/10.1016/j.compenvurbsys.2025.102335

🧾 BibTeX (click to expand)

@article{yang2025hyperlocal,
  title        = {Hyperlocal disaster damage assessment using bi-temporal street-view imagery and pre-trained vision models},
  author       = {Yang, Yifan and Zou, Lei and Zhou, Bing and Li, Daoyang and Lin, Binbin and Abedin, Joynal and Yang, Mingzheng},
  journal      = {Computers, Environment and Urban Systems},
  volume       = {121},
  pages        = {102335},
  year         = {2025},
  doi          = {10.1016/j.compenvurbsys.2025.102335},
  publisher    = {Elsevier},
  url          = {https://doi.org/10.1016/j.compenvurbsys.2025.102335}
}


## 🗂 Repository Structure

```bash
BiTemporal-StreetView-Damage/
│
├── codes/                          # Model training and evaluation scripts
├── images/                         # Project figures
│   ├── study_area_disaster_damage_made.png
│   ├── architect1.drawio (1).png
│   ├── design experiment.drawio (1).png
│   ├── dual_channel.drawio (2).png
│   ├── 0204-06.png
│   ├── readme.txt
├── LICENSE
├── README.md
└── .gitignore