Advancing Wildfire Detection: Using YOLOv8 on Multifarious Images

The escalating threat of forest fires, attributed to climate change and rising temperatures, underscores the urgency for advanced detection methods. This paper explores the effectiveness of employing the You Only Look Once object detection algorithm, specifically utilizing Bounding boxes to improve the model's performance by capturing texture variations between the fire and the background. The study aims to refine wildfire detection through the meticulous curation and annotation of a diverse image dataset. A comparative analysis of various YOLO models evaluates their performance. The dataset comprises both fire and non-fire images, showcasing diverse backgrounds to expose the model to real-world scenarios for comprehensive training. Despite the inherent challenges posed by variations in image textures and backgrounds, the models demonstrated consistency. The best-performing model achieved an mAP50 (mean Average Precision at 50% intersection over union) of 91%, while an unprecedented mAP50:95 of 68.5% further attested to the model's efficacy. These results present a significant advancement in wildfire detection capabilities.