现代电子技术

针对传统红外热成像甲烷气体泄漏检测中需要先验背景、易受特征相似干扰物影响，以及受实时环境因素影响特征变化较大的问题，文中提出一种基于YOLOv5s与红外热成像的甲烷气体泄露检测方法（GAS⁃YOLOv5s）。首先，使用K⁃means算法对YOLOv5模型的锚框进行预分类优化，根据泄漏点距离和泄漏程度确定锚框大小；其次，引入坐标注意力（CA）机制，提高算法对泄漏点位置附近气体特征提取的效率。实验结果表明，改进的YOLOv5s模型相较于原始模型，精确率（P）、召回率（R）、mAP@0.5和mAP@0.5：0.95分别提高了6.5%、1.9%、4.8%、1.6%，在视频检测集上达到263 f/s的检测速度，满足实时检测要求。证明改进的YOLOv5s算法提升了使用红外热成像甲烷气体泄漏检测的准确率，实现了甲烷气体实时多级分类检测，可以部署到各类检测终端。

The traditional infrared thermal imaging methane gas leakage detection needs prior background, and it is vulnerable to characteristically similar substance, and its characteristic change is significant under the influence of real⁃time environmental factors, so a methane gas leakage detection method GAS ⁃ YOLOv5s based on YOLOv5s and infrared thermal imaging is proposed. Firstly, the K⁃means algorithm is used to optimize the pre⁃categorization of YOLOv5 model anchor whose size is determined by shooting distance and leakage rate. Secondly, the coordinate attention (CA) mechanism is introduced to improve the efficiency of gas feature extraction near the leakage point. Experimental results show that the precision, recall rate, mAP@0.5, and mAP@0.5:0.95 of the improved YOLOv5s model is improved by 6.5%, 1.9%, 4.8%, and 1.6%, respectively, in comparison with those of the original model. Its detection speed can reach 263 f/s on infrared video test set, which meets the requirement of real⁃time detection. The results show that the improved YOLOv5s algorithm improves the accuracy rate of infrared thermal imaging methane gas leakage detection, and realizes real⁃time multi⁃level classification detection of methane gas, so it can be deployed at various detection terminals.