Fire and Smoke Monitoring with Alert System for Enhancing Forest Security

Abstract

Real-time forest fire detection requires the prompt and accurate identification of fire and smoke, enabling the timely generation of actionable alerts for verified detections. In this study, an approach to real-time detection of forest fires and smoke based on YOLOv11 vision model was proposed. It includes an efficient process of object localization, confirmation of detected events based on confidence scores, and generation of Short Message Service (SMS) notifications. The dataset of fire and smoke images from Kaggle was divided into the training, validation, and test sets with a proportion of 70:20:10 respectively. Optimized fire and smoke detection were performed on the annotated images. The proposed detector analyzes the stream of images and videos to produce class-specific bounding boxes and confidence scores, and verified detections lead to remote alerting. Experimentally, it results in 96.8% precision, 95.9% recall, 97.4% mAP, and 88 Frames Per Second (FPS).