Autonomous Firefighting Drones Using Vision Transformer Model for Smart Fire Detection Systems

This research introduces an innovative method for intelligent fire detection utilizing autonomous drones equipped with a Vision Transformer (ViT) model. The technology utilizes airborne monitoring and deep learning to detect fire outbreaks in real-time with improved precision and scalability. A dataset of 10,000 labelled aerial images, encompassing fire and non-fire events, was utilized to train and evaluate the model. The proposed ViT-based system achieved an overall accuracy of 96.4 %, markedly improving traditional CNN-based models such ResNet50 and MobileNet, which reached accuracy of 91.2 % and 88.7 %, respectively. The system exhibited a precision of 95.7 %, a recall of 97.1 %, and an F1score of 96.4 %, signifying dependable performance across various environmental conditions. Field testing involving autonomous drone deployment in three simulated forest areas demonstrated a real-time fire detection latency of less than 2.3 seconds. These findings highlight the capability of ViTintegrated drones for quick autonomous fire detection, facilitating expedited response and reduced risk to human life and property.