Awards & Nominations
IQ BIT has received the following awards and nominations. Way to go!
Automated Detection of Hazards
Smart Fire hazards mitigation tool
Summary
Develop a platform for taking decisions based on the hazard behavior predictions and the vulnerability index of nearby areas. It is divided in 3 stages:1- Use Convolutional Neuronal network to detect fire hazards and calculate from manipulated data the vulnerability index per area. 2- Predict behavior of a hazard throw a the ML model and determine fire propagation index based on weather, soil, water resources and field reports. 3- Recommend the most effective actions based on the simulated scenarios and the vulnerability index.We are handling a DEMO of stage 1 for Arkansas state and an interface to easily consume this information and manipulated the weight variable of the index.
How We Addressed This Challenge
We have developed a Machine Learning model based on convolutional neuronal network that analyzes NASA satellite images and performs image segmentation to identify the presence of smoke clouds to detect fire hazards. In addition, we have extracted and manipulated relevant data from Arkansas to calculate the vulnerability index per Arkansas county and determine the level of danger in nearby counties to the fire.
To easily consume this info we have developed a web app interface that shows the fire geo-localized and the vulnerability index per county.
As time is particularly important to safe lives, ecosystems, animals and plants, economic losses, environment pollution, among others; having a tool that allows decision makers to make decisions based on relevant information and simulation in time of the hazard, estimate the impact of these decisions and easily compare the effectiveness of each possible action becomes highly relevant.
It is also relevant to design mitigation strategies considering the vulnerability index per area and the simulate of possible fire hazard scenarios.
The ML model detects fire hazards and the app takes this information to display the fire information and calculate the vulnerability index of the nearby counties. The app shows a fire hazard and the vulnerability index by county. It is possible to manipulate the weight for each of the six variables (Soil humidity, population density, type of vegetation, topography, whater/ earth index) of the vulnerability index to prioritize the most and less relevant variables.
It is also possible to overlap layers with soil humidity, population density, topography, whater/earth index.
The information used to calculate the vulnerability index is public and can be found online, it is representative information for fire propagation and damage.
In order to make all the variables equally relevant, we have normalized all to a value from 0 to ten where 0 is none-vulnerable and 10 is highly vulnerable.
We aim to develop a platform that makes smarter decisions based on real data predictions of the hazard behavior and the vulnerability index of near areas. The development of this platform is divided in 3 stages:
- Stage one is about a model that detects fire hazards and calculates the vulnerability index per area. We have developed an interface that will permit this info to be easily consumed by decision makers. An alert will show the fire geo localization and will show the vulnerability index of near areas.
- Stage two is about to predict the behavior and development of a fire hazard, this will be done by developing a model to determine fire propagation index based on weather, soil, water resources, accessibility and real time field reports. This in conjunction with real time satellite images will adjust the convolutional neuronal network.
- This last stage is about to recommend the most effective actions to take based on simulated scenarios. This scenario will take into account a calculated attenuation fire index based on the type of possible actions and the vulnerability index of near areas.
How We Developed This Project
It is focused on Machine Learning, we have plenty fires near our city during summer and we believe that hazard management could be better and if we can help in some way, we are happy to do it.
The network follows a U-Net topology to localize smoke cloud areas by doing per pixel classification in the images. The output of the network is a probability on each pixel of whether it represents an area with smoke or not. The U-Net network is a specialized fully convolutional neural network that produces an output of the same dimensions as the input image, with a classification label for each pixel. In our case, this is a binary classification to determine if each pixel corresponds to a smoke cloud or not
Data analysis and algorithms can do a lot and we are getting interesting info to do it. Our approach was that for first time we heard about this satelital info that looks very interesting and we have the data experience. We don't have many possibilities to show our capabilities so its a nice opportunity. Our main approach was based on predictions and simulations.
Phython, R, pytorch, React, Leaflet, semantic UI.
a problem was reduced data quantity with label, even when we got a tool recommendation the labeling task was manual and no automatic.
We manage to work great as a team and support each other, after all we did generate a couple algorithms.
How We Used Space Agency Data in This Project
We used satelital images from Giovanni (NASA) specifically GLDAS NOAH 025 3H B2.1.
also we got DEM from SRTM90 and USDA
we used this data to estimate index vulnerability for each county in Arkansas respect to several variables such as: Soil humidity, population density, type of vegetation, topography, whater/ earth index.
Project Demo
https://drive.google.com/file/d/1VqWXu-uz7KBTfHh6YDSJfJrsLwCX4gfM/view?usp=sharing
Project Code
Data & Resources
We used satelital images from Giovanni (NASA) specifically GLDAS NOAH 025 3H B2.1.
also we got DEM from SRTM90 and USDA
we used this data to estimate index vulnerability for each county in Arkansas respect to several variables such as: Soil humidity, population density, type of vegetation, topography, whater/ earth index.