A Smoke and Fire Detection Model Based on Improved YOLO11

Smoke and fire detection plays a critical role in environmental monitoring and disaster prevention. However, due to the diverse and complex manifestations of smoke and fire, existing deep learning-based detection methods often suffer from insufficient feature extraction capabilities and high false positive rates. To address these issues, we propose a novel smoke and fire detection method called RSW-YOLO, which is based on the improved YOLOv11n. First, the structural reparameterization technology is introduced in the C3k2 module (RepC3k2), which enhances the model’s feature extraction capability and detection accuracy while maintaining efficient inference. Second, the SEGE attention is employed to suppress irrelevant background information and enhance key feature representations, effectively mitigating interference from false targets. Finally, applying the WIoUv3 loss function reduces the impact of adverse gradients caused by low-quality instances in the dataset. In evaluations on the D-Fire dataset, RSW-YOLO achieves $\mathbf{8 0. 8 \%}, 73.1 \%$, and 79.8% on P, R, and mAP@50, significantly outperforming existing methods. This research provides a new technical paradigm for smoke and fire detection in complex scenarios.