Wildfire Detection Using Computer Vision

Abstract

This work aims to enhance early and real-time wildfire detection utilizing computer vision and transfer learning techniques, specifically employing the VGG16 model. We developed two models, the first using only RGB images, achieving an accuracy of 88%, representing a 4% improvement over previously existing models. The second model utilize fusion technique, integrates both RGB and thermal images, attaining a remarkable 99% accuracy. Additionally, prototypes for future web and mobile applications have been created to facilitate real-time wildfire detection and response.