Disaster Damage Assessment

Inspiration

The increasing frequency of natural disasters worldwide has highlighted the need for efficient and data-driven disaster response strategies. Governments and emergency services often struggle to assess damage quickly and allocate resources effectively. Our project aims to bridge this gap by leveraging AI-powered image analysis to provide rapid and accurate disaster impact assessments. Trigger point was the recent wildfire in CA.

What it does

Our solution processes pre- and post-disaster images and generates a comprehensive damage assessment report. The report categorizes affected areas into high and low-priority zones, helping government agencies and emergency responders make informed decisions. By automating the assessment process, we aim to ensure faster and more efficient disaster response.

How we built it

We sourced pre- and post-disaster image datasets from publicly available resources such as the xBD dataset and LADI v2.
We developed a deep learning model that processes satellite and drone images to classify disaster severity.
The model employs convolutional neural networks (CNNs) for feature extraction and a classification network to determine the extent of damage.
We have partially implemented the core model but have yet to develop a user interface (UI) for deployment

Challenges we ran into

Finding high-quality, labeled pre- and post-disaster datasets was challenging.
Ensuring the model generalizes well across different disaster scenarios.
Computational constraints when processing high-resolution satellite imagery.
The lack of a fully developed UI to make the system easily accessible.

Accomplishments that we're proud of

Improved the model's ability to distinguish between high and low-priority zones.
Optimized preprocessing techniques for handling large-scale satellite imagery.

What we learned

What's next for Disaster Damage Assessment

Developing a web or mobile-based UI to make the system usable for end users.
Improving the model’s inference speed for real-time assessments.
Expanding the dataset to include more disaster types such as floods, wildfires, and earthquakes.
Deploying the model using TensorFlow Lite to enable edge-based disaster assessments.
Partnering with emergency response agencies to refine the model for practical applications.
Scaling the project to predict disasters before they occur by incorporating meteorological and geospatial data.
Enhancing the system to estimate the number of affected people and relay this data to nearby hospitals, enabling them to prepare in advance.