Oozma Kappa| Automated Detection of Hazards
Awards & Nominations
Oozma Kappa has received the following awards and nominations. Way to go!
Automated Detection of Hazards
AOTL - A Spatial-Temporal Deep Learning Predictor of Wildfires in US
Summary
Based on real-time spatial-temporal wildfire datasets from NASA, we have developed AOTL (Always On The Lookout) service based on RNN and CNN models to detect and predict (2 hours ahead) the possibility of wildfire spread in a geographic location, and provide visualization results to stakeholders in a readable version. To achieve our goals for human safety, ecosystem & infrastructure protection, we plan to provide interactive maps with alerts via SMS and email to local government bodies and civilians, which is our primary broadcasting strategy. We have also designed a website to present all our contributions.
How We Addressed This Challenge
We propose an Information and Communications Technology ICT-based service backed by hybrid-modeling approach that can help stakeholders take necessary actions based on following approaches:
- Early Prediction: build a model that will do the early prediction of wildfire based on the preceding data collected from that position and nearby places
- Human Safety: Intimate local government bodies and civilians within the geographic location of the occurrence of wildfire
- Ecosystem Protection: Alert US Fish and Wildlife Service FWS of an upcoming calamity
- Infrastructure Protection: Send automated call to 911 with the geographic location of the wildfire occurrence
Flow Diagrams:
Our state of the art model follows a two-fold multi-modal approach that includes:
Temporal Prediction of the occurrence of wildfire at hotspot locations within the next 2 hours anywhere in United States
Spatial Detection of the presence of fire in digital camera imagery
How We Developed This Project
Motivation
Every year, almost 6 million acres of land is burnt due to wildfires with more than 50,000 such wildfire occurrences on average within the United States alone. Though organizations such as Fire Information for Resource Management System FIRMS maintains daily and historical dataset of the geographic and temporal locations of wildfire, a centralized system for the dissemination of this information to all stakeholders is missing.
Our goal is to use state-of-the-art Deep Learning methods to present the early predictions and spatial detection in a form that is readable and helps facilitates the stakeholder in decision making.
Technology
- We use Python for Exploratory Data Analysis and model development
- Tensorflow Keras with Google Colab are used to develop our models.
- The UI is designed on Figma and hosted using GoDaddy free domain.
- To maintain team dynamics and chalk-out our work breakdown structure, we used Miro board, Slack and Google Meet.
Service-Architecture
How We Used Space Agency Data in This Project
We have used NASA MODIS C6 dataset collected from oct 26, 2020 to sep 3, 2020 to predict the wildfire before occurrence by RNN model based on previous and surrounding data.
Project Demo
Slides: https://www.slideshare.net/SaharaAli/oozma-kappa
User Interface: URL
Figma Prototype: URL
Website Prototype: URL
Project Code
Data & Resources
We have used following resources:
- “US year-to-date wildfire statistics from national interagency fire center.” https://www.nifc.gov/nicc/sitreprt.pdf, 2020.Accessed on October 4,2020.
- “Fire-smoke dataset from deep quest ai github repository.” https://github.com/DeepQuestAI. Accessed on October 4, 2020.
- “Modis c6 active fire data.” https://firms.modaps.eosdis.nasa.gov/activefire/firms−shapef ile,2020. Accessed on October4,2020.
- K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778, 2016