Optimization of YOLOv11 for Fire Hazard Detection in Ultra-Small Targets

Ancient buildings, as invaluable historical and cultural heritage, face severe fire prevention challenges, with incense burning identified as one of the primary fire hazards. Traditional fire source detection technologies often suffer from high false alarm and missed detection rates under complex conditions such as high humidity and heavy dust, making them unsuitable for high-precision monitoring required in heritage structures. With the rapid advancement of deep learning, image-based fire hazard detection methods have emerged as a promising research direction. This study proposes a lightweight fire hazard detection model based on an enhanced YOLOv11 architecture, specifically tailored for incense-burning scenarios in ancient buildings. The model incorporates three key components—the HWD module, the CCFM module, and the P2Head for small-object detection—to significantly improve detection performance for small fire-related targets while reducing computational complexity. Experimental results demonstrate that the improved YOLOv11 achieves increases of 2.6%, 1.4%, 1.4%, and 3.6% in Precision, Recall, mAP50, and mAP50–95, respectively. Although the frame rate (FPS) decreases by approximately 2.7%, it remains sufficient to meet the demands of real-time detection applications.