YOLOv8-Based Deep Learning Framework for Wildfire Detection in Remote Sensing Images

Abstract

This study proposes a fire detection method for remote sensing images based on YOLOv8 to improve the accuracy and real-time performance of fire target detection. Through training and testing on the FIRESENSE dataset, the experimental results show that YOLOv8 outperforms YOLOv5 and YOLOv7 in key indicators such as [email protected], [email protected]:0.95, recall rate and precision rate, showing stronger fire target recognition ability and false detection suppression ability. To further improve the detection effect, this study adopts data enhancement, small target detection optimization and efficient non-maximum suppression (NMS) strategy to improve the robustness of the model under complex backgrounds and different lighting conditions. In addition, through inference tests on different computing devices, the efficient detection ability of YOLOv8 in the GPU environment is verified, and it also has a certain edge computing adaptability. This study provides a high-precision, low-computing cost solution for intelligent fire monitoring systems, which can be applied to forest fire prevention, environmental monitoring and disaster warning, and provides important technical support for improving fire response efficiency.