Research on Fire Smoke Detection Algorithm Based on Improved YOLOv8

Fire has consistently posed a significant disaster risk worldwide. Current fire detection methods primarily rely on traditional physical sensors such as light, smoke, and temperature detectors, which often struggle in complex environments. The susceptibility of existing fire detection technologies to background interference frequently results in false alarms, missed detections, and low detection accuracy. To address these issues, this paper proposes a fire detection algorithm based on an improved YOLOv8 model. First, to enhance the detection capabilities for large-scale fire and smoke targets, a large target detection head is added to the backbone of the YOLOv8 model. This modification enhances the network's receptive field, allowing it to capture a broader range of contextual information and identify fires over extensive areas. Secondly, an efficient multi-scale attention mechanism, EMA (Efficient Multi-Scale Attention Module), based on cross-space learning is integrated into the FPN (Feature Pyramid Network) part of the model. This mechanism highlights target features while suppressing background interference. Additionally, a PAN-Bag (Path Aggregation Network Bag) structure is proposed to help the model more accurately detect objects such as fire and smoke, which have uneven feature distributions and variable morphologies. With these improvements, we introduce the YOLOv8-FEP algorithm, which offers higher detection accuracy. Experimental results demonstrate that the YOLOv8-FEP algorithm improves the mAP by 3.1% and the accuracy by 5.8% compared to the original YOLOv8 algorithm, proving the effectiveness of the enhanced algorithm.