- PI: Guy Schumann, ImageCat
- Co-I: Marina Mendoza, Krasniansky/ImageCat
- Co-I: Ron Eguchi, ImageCat
- Co-I: Ajay Gupta, HSR
- Co-I: Unmesh Kurup. Intuition Machines
- Co-I: Ron Hagensieker
Machine Learning (ML) brings significant advances in the area of disaster impact mapping and predictions. However, the biggest challenge facing ML and Earth Observation (EO) is that of truth labeling, which an ML model cannot get enough of to perform its tasks as accurate as possible. This is particular challenging when faced with extreme and relatively infrequent events such as floods and wildfires. As a solution, our project team proposes to use the famous hCaptcha service to collect millions of labels of wildfires by the global web community. The service protects websites and privacy by running a simple test to distinguish humans and robots.
Captcha is therefore extremely powerful to collect millions of labels that can be used to develop and improve ML models for a large range of applications. From these very large number of labels, we can train a much more powerful ML model for real time detection of fire and smoke intensities. Our team comprises experts in EO and ML as well as experts working with the healthcare sector.
This combined expertise will allow us to assist response teams in preventing disaster-related health issues, such as symptoms linked to wildfire smoke inhalation that are of growing concern and can affect all age levels of the population. Having access to real-time maps of smoke cover or affected areas would, for example, enable quick and smart re-routing of population in order to avoid the smoke-affected air and its potentially harmful consequences.
Related Impact
