Fire Detection Algorithm Based on Improved YOLOv8

Fire identification plays a crucial role in the realm of fire safety, serving as a vital component for the early detection and swift response to fire incidents. Effective fire identification is essential in minimizing the damage and losses associated with fires, as it enables timely interventions. we introduce a novel fire detection method that employs RHE-YOLOv8, aiming to enhance the accuracy of fire identification and localization. By leveraging this advanced approach, we seek to optimize the response speed of fire identification systems, ultimately facilitating more precise recognition and pinpointing of fire sources. In this study, the YOLOv8 model incorporates the RepGhost module to decrease its floating-point computations, thus achieving a more lightweight structure. Furthermore, the seamless integration of the Hyb-ECA attention mechanism into the network's neck structure has notably boosted the model's recognition accuracy, all while maintaining a low computational footprint. Experimental findings demonstrate that these enhancements can elevate the average precision (mAP) of the YOLOv8 model by 2.5%. Simultaneously, the model undergoes a decrease in floating-point operations (GFLOPs) of 13.6%. These enhancements significantly improve both the precision and effectiveness of fire detection, demonstrating their efficacy in real-world fire situations.