考虑不确定性的红外视频气羽分割及其在VOCs泄漏检测中的应用  

Infrared Video Plume Segmentation Considering Uncertainty and Its Application in VOCs Leak Detection

作　　者：张宇霖 谷小婧[1] 顾幸生[1] ZHANG Yulin;GU Xiaojing;GU Xingsheng(Key Laboratory of Smart Manufacturing in Energy Chemical Process,Ministry of Education,East China University of Science and Technology,Shanghai 200237,China)

机构地区：[1]华东理工大学能源化工过程智能制造教育部重点实验室,上海200237

出　　处：《石油学报(石油加工)》2025年第1期307-318,共12页Acta Petrolei Sinica(Petroleum Processing Section)

基　　金：国家自然科学基金项目(61973122、62173143)资助。

摘　　要：针对红外气体成像实现泄漏检测自动化存在的安全问题,提出一种可同时进行视频气羽分割与不确定性估计的挥发性有机物(VOCs)泄漏检测框架。通过分析视频帧判断泄漏位置与范围,并将深度模型的不确定性结果作为检测结果的可靠性指标,为后续决策提供参考,以增强算法使用中的安全性。通过记忆模块来关联视频时序信息,实现快速高效的视频气羽分割与不确定性评估,提升了检测结果的准确性和连续性。为缓解VOCs视频数据稀缺问题,使用真实与合成数据混合训练。结果表明,该方法在给出泄漏检测结果的同时,可以输出模型置信度评价,避免对深度模型的盲目依赖,为人工决策和修复方案的制定提供更多依据。A VOCs leak detection framework that simultaneously performs video plume segmentation and uncertainty estimation is proposed to address the issues of automated leak detection by infrared gas imaging.The location and range of leaks can be determined by analyzing video frames,and the uncertainty results of the depth model are used as a reliability indicator for detection results.This provides a reference for subsequent decision-making,thus enhancing the security in use of algorithm.Memory modules are used to interconnect video temporal information,by which fast and efficient video feather segmentation and uncertainty evaluation have been achieved,thus improving the accuracy and continuity of detection results.To alleviate the scarcity of VOCs video data,a mixture of real and synthetic data is used for training.The experimental results show that this method can simultaneously output leak detection results and model confidence evaluation data,which can avoid blind dependence on deep models,and provide more evidences for manual decision-making and the formulation of repair plans.

关 键 词：视频语义分割 气体泄漏检测 不确定性估计 合成数据 

分 类 号：TE624[石油与天然气工程—油气加工工程]