An Improved Fire Detection Model Based on YOLOv11n

Fire detection technology plays a critical role in the field of fire safety. However, conventional smoke detector-driven fire alarm systems frequently exhibit limitations in reporting fire incidents promptly and effectively. To address this limitation, this study proposes FD-YOLOv11, an enhanced fire detection algorithm based on YOLOv11n. Specifically, a parallel patchaware attention (PPA) module is introduced to augment small target detection capabilities, thereby enhancing the model's object detection accuracy. Furthermore, the convolutional kernels of a lightweight deep neural network, termed DualConv, are integrated. This integration not only strengthens the model's detection performance but also reduces its parameter count and computational requirements. Additionally, a Histogram Transformer Block (HTB) is incorporated into the architecture, significantly improving the model's capacity to maintain highquality detection results in smoke-obscured environments. Comparative evaluations between the enhanced YOLOv11 model and existing state-of-the-art models were conducted. Experimental results demonstrate that the proposed model exhibits superior detection performance compared to the baseline model, achieving an increase of 3.2% in mAP@50 and 1.2% in mAP@50:95.