Vision Transformer based Forest Fire Detection for Smart Alert Systems

A forest fire is a vast, devastating fire that quickly spreads over brush or woodland. Globally, the number of forest fires has drastically increased during the past few years causing massive damage to the environment and human life. Here, we propose an intelligent system that can help in the early detection of forest fires before they spread and cause massive damage. The proposed system uses state-of-the-art Vision Transformer (ViT) architecture for Forest Fire Detection. The system takes a video feed as input and feeds it into the model for detection. Once the model detects a fire or any potential indication of fire, emergency alerts are sent out through SMS, Email, and WhatsApp. Further, the YOLO v3 algorithm has been used for Human Detection from the images in which fire was detected. If any humans are detected, the system sends the latitude and longitude of that particular location along with a Google Maps URL to the Forest Department officials or the concerned authorities via alerts, so that search and rescue can be performed swiftly. The proposed fire detection model yields an overall accuracy of 98% with an F1-Score of 0.98, while the YOLO based human detection model is 83% accurate on video test data.