The solution utilizes a convolutional neural network and stochastic gradient descent. It achieves a final validation accuracy of 83.82% and a final AUC score of 86.11%.
The dataset used in this project was Kaggle Wildfire Detection Image Data.
The model is a part of a greater solution, which encompasses the use of hardware microcontroller and cameras to receive raw data from the surrounding environment.
Wildfires are dangerous natural disasters. They destroy animals habitat, kill animals, destroy human houses, and displace humans as “refugees” of natural disaster. Although they may be helpful by removing diseases, the increase in frequency of them due to climate change and global warming has proven to make them more catastrophic than helpful to the environment. The question that we will have as we continue is as below: how can we use machine learning technology to effectively detect wildfires based on pre-existing data?
The main purpose of this solution is to alert fire lookouts early on about fires. There are not sufficient number of fire lookouts to make sure that there is no fire in every meter square of the natural environment. For that reason, they could use this solution to aid the process.
Please visit FireNet Webpage for more info on the background research.