WILDFIRE DETECTION SYSTEM USING DEEP LEARNING ALGORITHMS

ABSTRACT

In light of the growing prevalence and severity of wildfires, the shortcomings of conventional fire detection methods have become increasingly apparent. This research presents an advanced approach that leverages the power of deep learning algorithms to specifically address the challenge of wildfire detection and classification. The Flame Vision dataset is a robust and comprehensive collection of wildfire images that was leverage on and the dataset undergone preprocessing, visualization and augmentation in order to enhance its quality and diversity. Employing the Flame Vision dataset, this study scrutinizes three cutting-edge deep learning models namely the Artificial Neural Network (ANN), the Convolutional Neural Network (CNN), and a transfer learning framework utilizing Visual Geometry Group (VGG-16). Each model undergoes extensive training and validation, with performance evaluation metrics on the test set in order to determine the most effective algorithm for wildfire detection and classification. In conclusion, the results show that transfer learning models using VGG-16 offer the most promising results for the task of wildfire image classification. The VGG-16 outperformed CNN and ANN in accuracy with 97.22%, 97.00% and 88.22% respectively and in F1-score with VGG-16 as 0.9429, CNN as 0.9299 and ANN as 0.3244. With these results, it can be concluded that the VGG-16 is a better classifier for the Flame Vision dataset.