High-accuracy combustible gas cloud imaging system using YOLO-plume classification network

Effective natural gas leakage detection is of great significance in terms of economy, environment and safety. Due to the irregular shape and ambiguous boundary of the gas, traditional motion detection algorithms are difficult to adapt to the changes in the gas movement state with the environment, resulting in an increased probability of false alarms. To address this issue, this paper proposes a gas plume-constrained YOLOv11 model based on infrared imaging detection technology, named YPCN (YOLO-Plume Classification Network). A new backbone feature extraction network, MobileNetV4, is selected to replace the original backbone network, and SPD-Conv is introduced in the segmentation head network. This network effectively reduces model complexity and enhances inference speed while maintaining detection accuracy. Additionally, a gas plume model is introduced as a key physical constraint condition in the loss function to enhance the model’s accuracy, segmentation precision, and generalization ability in handling gas plume tasks. Moreover, this paper constructs a gas leakage dataset consisting of 13,109 frames, covering different distances, sizes, and backgrounds. Experimental results show that the proposed model achieves an F1-score of 88.97% and an IoU of 89.74%, improving upon the baseline by 7.37% and 7.59%, respectively, with a detection accuracy reaching 99.78%.