EMO-YOLO: an enhanced model for cross-scenario fire and smoke detection

To address the issues of insufficient accuracy, poor real-time performance, and limited application scenarios in existing fire and smoke detection models under complex environments, this paper constructs a cross-scenario fire and smoke dataset and proposes a lightweight real-time detection network model incorporating EMO-YOLO. Firstly, the model reduces parameter size by integrating the lightweight EMO network with the backbone network of YOLOv5s.70. Secondly, the CoordAttention mechanism is introduced to enhance the fine-grained feature extraction capability for smoke, flame, and human targets. Experimental results demonstrate that the improved model achieves a mean average precision (mAP@50) of 91.4%, a precision rate of 99%, with a model size of only 4.26MB. It achieves an inference speed of ≥30 frames per second on edge devices, enabling simultaneous and accurate identification of smoke, flames, and personnel in target areas.