After hours of brainstorming and going through various potential solutions, we finally had the solution that we thought would be efficient, optimal and can work. The picture was quite clear to us, creating an IoT device that would read the data from nature and use it to predict whether there is going to be a forest fire, display that data and the prediction on a dashboard. If a forest Fire is predicted, a certain Forest Fire mitigation system must be activated.
IoT is the most essential part of our solution, we wanted to design a device that would have the appropriate number of sensors so it could both detect early forest fires and have as small carbon footprint as possible by consuming less energy. Looking at these needs, we designed a concept, which can be seen in the following image. For the scope of this Prototype Sprint, we decided to go for a prototype version of the concept that would have only the basic and essential sensors. The sensors the prototype had are:
-
Air Temperature and Humidity sensor (DHT11)
-
Capacitive Soil Moisture sensor
-
Soil Temperature sensor (DS18B20)
-
Sunlight Sensor (SI114X)
 |
 |
|---|---|
| Yaksha Concept | Yaksha Prototype |
The IoT device was coded in Python. The Yaksha device takes new readings every minute. Each minute is divided into 2 parts, the polling time and the sleep time, each of which is 30 seconds long. During the polling time each sensor takes 10 readings, and the average is returned as the final reading for that minute. The result is then sent to the Yaksha Dashboard (currently, IoT Central) and is also used to predict the forest fires, if a forest fire is predicted, the Forest Fire Mitigation system or the “Forest Fire First-aid" is deployed. The next 30 seconds is the rest phase for the device. The sleep time can be increased or decreased as per the requirement.
 |
|---|
| Block Diagram of Yaksha Prototype |
The forest fire detection itself was not sufficient for us, we needed to go beyond that and find a way in which we could mitigate or prevent forest fires. For the first fire mitigation system, initially, we were thinking of using drones that would carry the water and sprinkle it on the affected area, but that turned out ineffective as there was only a limited amount of water that a drone could carry. So, we pivoted to a fireball, which on contact with fire or around certain degrees of temperature, will automatically burst open and within a few seconds extinguish a fire. This solution seemed very promising as these balls are small, and a drone can carry many such balls. We named this Forest fire mitigation system the “Forest Fire First-aid".
 |
 |
|---|---|
| Fire Extinguishing Ball | Drone carrying the balls |
For the dashboard of the IoT device, the team decided to utilize Azure IoT Central. The primary reason is because it speeds up the development of the dashboard and makes it easier to connect to the IoT device. Aside from ease of use, IoT Central provides a secure workspace for the team. Featuring industry-leading security, the only thing left for the team is to specify which accounts can access the dashboard and the team can now make a simple, secure, functional, and informative dashboard.
 |
|---|
| Yaksha Dashboard |
After live forecasting from Yaksha device, it is utmost important to
predict the early wildfire to take caution to prevent the fire by the
concerned authorities based on the warning generated by Machine Learning
Model which means it is key component towards forest-fire control.
This model will be trained using meteorological dataset based on the
specific geographical location which considers several parameters like
temperature, humidity, soil moisture, geographical coordinates, etc.
Machine Learning approaches used in wildfire detection can lead to potential outcome with better accuracy to mitigate damage caused by fire-
- Fire detection and Mapping
- Occurrence of potential risks
- Fire behavior and effects
The proposed system will utilize the meteorological parameters to perform computation using algorithms like Logistic Regression, Random Forest Classifier, K-nearest Neighbors, Support Vector Machine, Decision Tree classifier, etc. which can lead to potential model with the highest accuracy to predict early fire.
 |
|---|
| Proposed Architecture for ML Model |
The following architecure diagram shows the complete working of Yaksha. This is a proposed model and is not a part of the prototype sprint.
 |
|---|
| Yaksha Architecture Diagram |
*Note: The Forest Fire First-aid and the Machine Learning model are two components that are currently incomplete as they require more research and we considered them out of the scope of the Prototype sprint while working on this project.