An AI-based Image Recognition System for Early Detection of Forest and Field Fires

Abstract

Forest and field fires have severe global implications, causing significant environmental and economic harm. Traditional methods of fire detection often rely on human personnel, which can pose safety risks and reduce their efficiency in large-scale monitoring. To address these challenges and minimize losses, there is an urgent need for real-time fire detection technology. In this research, we propose the utilization of artificial intelligence techniques, specifically Deep Learning with Convolutional Neural Networks (CNN), to tackle this issue. Our proposed system analyzes real-time images captured by IP cameras and stored on a cloud server. Its primary objective is to detect signs of fires and promptly notify users through a mobile application, ensuring timely awareness. To train our model, we meticulously assembled a dataset by merging three existing datasets comprising both fire and non-fire images. Additionally, we incorporated images that could potentially be misinterpreted as fire, such as red trees, individuals wearing red clothing, and red flags. Furthermore, we supplemented the dataset with images of unaffected areas obtained from online sources. The final dataset consisted of 1,588 fire images and 909 non-fire images. During evaluations, our model achieved an accuracy of 93.07%. This enables effective detection, thus rapid intervention and damage reduction. It is a proactive and preventive solution to combat these devastating fires.