GasEdge-YOLO: multi-scale edge generator and star-shaped branch attention for efficient infrared gas detection

Abstract Industrial invisible gas leakage detection is critical for environmental protection and hazard warning, yet automatic detection in infrared images faces challenges like low contrast, weak edge features of gas targets, and complex background interference. To address these issues, we propose GasEdge-You Only Look Once (YOLO), an enhanced detection model built on YOLOv11, integrating two novel modules: the multi-scale edge generator (MSEG) and the star-shaped branch attention block (SBAB). The MSEG module strengthens directed focusing and semantic enhancement of gas edge information through an additional multi-scale edge feature extraction branch, effectively capturing blurred gas boundaries. The SBAB, embedded in the C3k2 module with the innovative multi-branch context anchor attention, efficiently fuses global contextual information, balancing detection performance and computational efficiency. Experimental results on the benchmark InfraGasLeakDataset demonstrate that GasEdge-YOLO outperforms YOLOv11-n by 5.5% in mAP50, significantly surpassing state-of-the-art object detection networks. This work validates the effectiveness of the proposed method, providing valuable insights for hazardous chemical gas leakage detection.