Forest Fire Detection

Project Title

Forest Fire Detection

Project Area of Specialization Artificial IntelligenceProject Summary

Many systems have been designed such as sensor-based systems, wireless systems for the detection of forest fire.The main problem with the sensor-based system is its cost and maintenance. 
So,working for the development of sensor less systems has started, which requires no/less power consumption, least maintenance and least cost with high accuracy and early detection.  
We are using the deep learning neural networks for designing the model, only needs high resolution cameras for the early detection of fire. 
Another main goal of our project is to reduce the complexity of the model to make it more faster than the previously existing models, which results in early detection. We are using Inception V3 for our project ,which is the 3rd version by Google in the deep learning neural networks.It is trained on the dataset of thousand classes taken from the ImageNet dataset, which is trained on more than one million training images.
 We are using the transfer learning technique.Smoke is an additional feature,we are considering in our project. Our model  is giving 92.3 percent of accuracy on 3000 images.
 A high-resolution camera which will take a Realtime video feed and uses our proposed model which has the reduced computational complexity and high speed of detection and will detect fire.

We will make it a product by interfacing NVIDIA Jetson developer kit with our camera.
When we will finish this task, our next goal is deforestation detection in forests, which we have included in the extended scope of our project.

Project Objectives

Our objective is to design the Convolutional Neural Network based architecture which: 
•    Reduces the computational complexity 
•    Will be a cost effective and gives accurate detection 
•    Real time demonstration of implemented model

Conversion of our project into a product by using NVIDIA jetson developer kit.

Project Implementation Method

Procedure consists of following steps:    
i.    Selection of pre-trained deep CNN-model, data gathering and its preprocessing 
ii.    Splitting of dataset into training and testing. 
iii.    Training of the model by using the preprocessed data 
iv.    Real time Testing and performance evaluation. 
v.      Interfacing with NVIDIA board for conversion into a product

Benefits of the Project

Our product would be deployed in forests which will early detect fire ultimately reduce the risk of wildfire and destruction.

Technical Details of Final Deliverable

We are designing the deep learning based neural network by using the transfer learning technique.Which we are retraining and modifying for our task i.e. for fire detection and will test on the testing dataset. 
For real-time demonstration, we will deploy a camera which takes the live video feed and use our proposed model to detect fire.For conversion into a product form,we will use NVIDIA jetson developer kit,which includes interfacing of kit and camera.
So we will have a complete product at the end.

Final Deliverable of the Project HW/SW integrated systemCore Industry OthersOther Industries Education Core Technology Artificial Intelligence(AI)Other Technologies OthersSustainable Development Goals Climate Action, Life on LandRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	46000
NVIDIA Jetson Developer kit	Equipment	1	26000	26000
Digital camera	Equipment	1	20000	20000