Real-time wildfire-prone area monitoring and early warning system based on two-stage YOLO-based smoke detection model

Wildfires present a severe risk to environmental sustainability, public health, and economic stability, especially in wildfire-prone areas such as northern Thailand. This thesis proposes a real-time wildfire-prone area monitoring and early warning system designed to address these challenges. First, a web-based application was developed to integrate IoT cameras with a YOLOv5-based smoke detection model for continuous monitoring and real-time notifications. Second, a two-stage smoke detection framework is introduced, leveraging Gaussian filtering and a dual-stage YOLOv5 pipeline to reduce false predictions and enhance detection reliability. The system was trained using the FireSpot dataset, comprising annotated images of early-stage wildfires, and demonstrated significantly improved performance over baseline models. Deployed in three municipalities in Chiang Mai, Thailand, the system has proven effective in delivering timely alerts and supporting wildfire management efforts. By integrating IoT technologies with machine learning techniques, this work provides a scalable and practical solution for early-stage wildfire detection, contributing to mitigation of environmental and public health impacts.