A Lightweight Fire Detection Framework Based on YOLOv11 for Deployment on Embedded UAV Systems

Forest fire poses a significant global threat, endangering ecosystems, biodiversity and the harmony of natural habitat. As the forest fires becomes more frequent and intense, the loss of natural habitat has increased rapidly. Therefore, it becomes important to detect forest fire at an early stage so that prompt and effective measures can be implemented immediately. To address this, we propose a low-cost drone-based forest fire detection and monitoring system that utilizes computer vision and deep learning for real-time forest fire detection. The proposed system uses a custom-built Quadcopter, equipped with the Raspberry pi camera module, to capture the real-time feed. These feeds are analyzed by the YOLOv11n object detection algorithm to accurately detect the fire. The proposed work utilizes an open-source dataset from Roboflow which contains 3426 images. This developed system is tested in various controlled environments demonstrating high mAP score of 93.6% in detecting wildfire. The YOLOv11n model achieves an accuracy of 92% and an approximate 8 frames per second for the test experiment. Therefore, it can be an effective tool for early wildfire detection, notifying timely alerts and aiding in rapid response, which can significantly enhance wildfire prevention and control efforts.