Assessing the Effectiveness of Object Detection Models for Early Forest Wildfire Detection

Wildfires, sometimes known as forest fires, are fast-moving, uncontrolled blazes that sweep through forests, grasslands, and other vegetation. Dry weather, high temperatures, and combustible vegetation all contribute to these devastating catastrophes, which result in significant ecological, economic, and human losses. Early detection is critical for mitigating these consequences, but existing wildfire detection systems are frequently reactive, relying on signs such as visible smoke or post-fire evaluations, limiting their capacity to identify flames in their early stages. This work addresses the need for a more proactive and efficient strategy for wildfire detection by comparing the performance of four sophisticated object identification models: Faster RCNN, MobileNet v2 SSD, YOLOv8, and YOLOv9. These models were trained using a customized dataset of 2,321 photos depicting early fire signs, such as tiny smoke plumes and hotspots. Each model was fine-tuned to increase its mean Average Precision (mAP) score in order to improve continuous monitoring and early detection capabilities. The findings show that YOLOv9 had the greatest mAP of 88.5%, followed by YOLOv8 with 80.50%. Faster RCNN and MobileNet v2 SSD also effectively detect tiny fire signs. This research helps reduce the devastating potential of forest fires while improving public safety by allowing for faster and more accurate wildfire detection.