关键词: YOLOv11n, 火灾烟雾检测, 小尺度目标检测, 轻量化网络, 计算冗余

Abstract: To address the challenges in edge-based fire and smoke detection, such as difficulties in identifying small-scale targets, low detection accuracy in complex scenes, and computational redundancy, a lightweight detection algorithm based on an improved YOLOv11n is proposed. First, a PMSDA attention module is designed to enhance the model’s perception of small-scale targets. Second, an ADown downsampling module is adopted to replace the original convolutional structure, significantly reducing the number of parameters and computational cost. Additionally, a lightweight and efficient dynamic upsampler named DySample is introduced to increase the spatial resolution of feature maps and enhance the representation of fine-grained features. Finally, a Compact Inverted Block is integrated to optimize parts of the deep layers, effectively mitigating redundant computation. Experimental results show that the improved model achieves a mAP@0.5 of 91.3%, a precision of 91.1%, and a recall of 89.5%, representing increases of 1.7, 0.5, and 1.1 percentage points, respectively, compared with the baseline model. The parameter count and GFLOPs are reduced by 25.3% and 20.6%. Deployment tests further demonstrate that the model achieves an inference speed of 102.8 FPS on GPU and 38.6 FPS on embedded devices, while maintaining a mAP@0.5 of 84.8%, satisfying the real-time detection requirements of fire and smoke monitoring on edge platforms.

Key words: YOLOv11n, fire and smoke detection, small-scale object detection, Lightweight Network, computational redundancy