Flame Detection Based on Deep Learning

This article describes the development of a real-time flame detection system employing deep learning technology. The YOLOv8 algorithm, which is well-known for its great speed and accuracy in object detection, is employed. The YOLOv8 model undergoes training using flame images and video data sourced from various scenarios, followed by testing on an Nvidia Jetson Nano device. This compact computing device is well suited for edge AI applications, addressing resource and energy limitations in high-risk environments. Experimental results show that YOLOv8, especially the YOLOv8m variant, has a maximum accuracy of 87.3% and an F1 score of 75% at a confidence level of 59.9%, a good balance for flame detection. Additionally, YOLOv8n achieves approximately 8-15 FPS at 320x320 resolution on the Nvidia Jetson Nano, making it ideal for real-time edge AI applications. This system operates efficiently on the Nvidia Jetson Nano, enabling real-time flame detection with precision, making it ideal for deployment in hazardous environments that demand adaptability. This study underscores the promise of employing deep learning technology to create an efficient and rapid response flame detection system in dangerous environments.