RoBoun| Spot That Fire V3.0
Spot That Fire V3.0
Artifical Intelligence Asisted Fire Detection System- Codename Flameseeker
Summary
The goal of our artificial intelligence-supported Python and Arduino-based solution system called Flameseeker is to detect fires as soon as possible and minimize fire damage.A fire detection program that prevents even bush fires from being overlooked, controls conditions where there is a high risk of fire, alerts authorities,and monitors data such as temperature, humidity, smoke, wind 24/7.
How We Addressed This Challenge
We have developed an AI supported, Python and Arduino based fire detecting system. 90 percent of fires start from the ground, spread and then rise; instead of traditional watchtowers, our system, which we designed at a height of 1 meter to the ground, can detect fires much earlier. Our project instantly compares weather and fire data from NASA and Meteomathics with possible risk parameters using artificial intelligence. It then alerts watchtowers when it detects danger with 88% accuracy in weather conditions. Active Arduino sensors, which we have already placed within the forest at an appropriate distance, are critical at this point. Officials at the Watchtower can check the data of these sensors from the phone app every second when they receive a warning, and in addition, our Arduino network can directly inform the fire department when it sees unusual events such as flames, smoke or sudden temperature increases. We promise that with our project; fires as devastating as the Australian bushfires in January and the Sacramento fires that are yet to be contained , which caused irreversible ecological and economic damage in addition to deforestation due to their late detection, will not recur.
How We Developed This Project
The sadness of our losses in Sacramento and especially the Australian fires led us to choose this competition. In order to get more detailed information about the general problems experienced in fires, we have interviewed the authorities and considered improving the fire detection used in their directive.
We used Adobe Xd and Flutter to desing and create our application. Firstly, we designed UI thanks to Adobe Xd, than by using flutter, we finished our application.In order to achieve a long-lasting and effective design that is resistant to harsh environmental conditions, we have experimented with SolidWorks and obtained the most suitable capsule.We used ardunio to collect and connect data from outside with the help of various sensors.In this project's data science and machine learning part we used python and python's publicly available libraries. We used;
-meteomatics.api library to get data from meteomatics database using coordinate and time data,
-pandas library to read csv files, simply look at how data structured and process the data if needed
-matplotlib and seaborn library to make plots and graphs to really understand the data
-scikit-learn library to make machine learning models and train those models with our processed data
Due to lack of budget, we had the problem of not being able to prototype the drone support we wanted to add to our system. In our experiment, we observed that our fire, humidity, smoke and temperature sensors worked extremely successfully. From our artificial intelligence-based application, we found that changes in the environment are displayed instantly in our application.
How We Used Space Agency Data in This Project
In this project we used data from NASA's Fire Information for Resource Management System (FIRMS)[1], UCI Machine Learning Repository[2], Wikipedia[3] and Meteomatics[4].
Our first approach was to use already preprocessed data in order to check if our problem solving method is suitable for this problem. For this reason we used Forest Fires Data Set from University of California Irvine's Machine Learning Repository[5]. There is an academic paper[6] analyzing and making a machine learning model using this data. We recreated the steps[7] to check how a model which trained using only meteorological data will perform. After our first approach we concluded that it is possible to detect wildfires using only meteorological data.
Secondly we wanted to see a connection using a small dataset of our own. We know that NASA already has a really well organized data about wildfires but our goal in this step was not to finalize our solution but to just check how our solution would work in some recent fires. We randomly selected different wildfires from 2020 California Wildfires Data[8] and prepared our data using python.
We wanted to check if we could find a clear connection between fire data and meteorological data. We used time and coordinate data from wikipedia to get meteorological data from Meteomatics.
Because we did not wanted to make this step too complicated we just took the temprature and humidity data. Using this data we created plots and graphs to really understand the connection.
We saw that in nearly 7 out of 8 cases humidity drops and temprature rises clearly before a wildfire.
Our final approach was to use data NASA's Fire Information for Resource Management System (FIRMS). With NASA's publicly available data[9] we can easily get all the wildfires aroud the globe.
We processed that data using Python's Pandas library and used that processed data to retest our solution method. We recreated the steps we took in our second approach and got data from meteomatics with our python script.
We used these data to check our method one last time and we concluded that our method indeed works as expected
Project Demo
Project Code
Data & Resources
[1] https://firms.modaps.eosdis.nasa.gov/
[2] https://archive.ics.uci.edu/ml/index.php
[3] https://en.wikipedia.org/wiki/Main_Page
[4] https://www.meteomatics.com/en/
[5] https://archive.ics.uci.edu/ml/datasets/forest+fires
[6] http://www3.dsi.uminho.pt/pcortez/fires.pdf
[8] https://en.wikipedia.org/wiki/2020_California_wildfires
[9] https://firms.modaps.eosdis.nasa.gov/active_fire/#firms-txt