Fire Smoke Detection Based on YOLOv11 Multi-Scale Fusion and Channel Aggregation

In environments like forests, parks, and schools, where fire and smoke are often small-scale, surrounded by complex scenes and obscured by objects, traditional detection methods fall short of real-time monitoring needs. This paper presents an innovative multi-scale fire and smoke detection algorithm based on YOLOv11 to tackle these challenges. The backbone network of YOLOv11 is restructured by incorporating the MOGA (Multi-Scale Difference Fusion) module, which enhances the model's capability to detect smoke and fire targets across a range of scales, thereby improving the detection accuracy for smaller fire and smoke incidents. Additionally, a Channel Attention (CA) module is introduced in the neck network to replace the upsampling process, enhancing the model's capability to detect objects with ambiguous boundaries. This modification reduces the loss of spatial information caused by upsampling, ultimately improving the accuracy of the YOLOv11 model. The improved model is evaluated using the D-Fire dataset. Experimental results demonstrate that, compared to the original YOLOv11 model, the enhanced version shows superior performance, with accuracy and mAP50 increasing by 1.2% and 1.4%, respectively, validating that the proposed algorithm significantly enhances the effectiveness of fire detection.