基于深度光流网络的光学气体成像中气体流速估计  

Estimation of Gas Velocity in Optical Gas Imaging Based on Deep Optical Flow Network

作　　者：顾笑宇 谷小婧[1] 丁杰[1] 顾幸生[1] Gu Xiaoyu;Gu Xiaojing;Ding Jie;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：《光学学报》2024年第18期312-322,共11页Acta Optica Sinica

基　　金：国家自然科学基金(61973122,62173143)。

摘　　要：聚焦于提升光学气体成像技术中气体流速估计的精度。首先,提出一种基于物理仿真的气体光流数据生成方法,构建真实且多样的气体光流数据集。此外,针对气体运动的特点设计一种基于梯度的损失函数,增强网络对气体边缘和内部微小运动的关注。最后,使用所提方法在合成数据上定量评估气体速度。结果表明,在气体光流估计任务中,经气体光流数据集微调的光流网络的性能提升较为显著,真实红外图像上的测试进一步证实所提方法在真实场景中的有效性。定量分析表明,使用所提方法估计得到的气体截面速度场准确率提升22百分点。所提方法不仅能够提高光学气体成像技术中气体流速估计的精度,也为类似应用的深度学习模型提供重要的技术支持。Objective Optical gas imaging(OGI)technology represents a non-contact method for detecting gases based on the distinctive infrared absorption characteristics.Compared to traditional contact gas detection methods such as catalytic combustion,electrochemical sensors,and semiconductor gas sensors,OGI offers advantages including a broad monitoring range,rapid response,high safety,and operational flexibility and efficiency.Over the past decade,it has been widely used in gas leak detection.Despite its success in detecting gas leaks,OGI technology encounters challenges in achieving high-precision gas flow rate measurements.Recent studies have attempted to address this issue by integrating optical flow algorithms with OGI technology to measure gas flow rates.These approaches estimate the optical flow of the leaking gas image using conventional optical flow algorithms and then calculate the gas velocity based on the ratio of pixels to physical length.However,the accuracy of this method remains limited in complex dynamic scenarios.With advancements in neural networks,scholars have explored the use of optical flow neural networks for gas flow estimation.However,existing optical flow neural networks are primarily designed for rigid body optical flow,and thus require significant modifications to address the unique physical and motion characteristics of gases.This necessitates meticulous adjustments and optimizations at each stage,from dataset construction to network design,to accommodate the special scenario of gas optical flow estimation.Methods We introduce a method for constructing a dataset of gas leakage optical flow using physics-based simulation software.Initially,the raw methane motion data are generated using the fire dynamics simulator(FDS).Subsequently,the three-dimensional gas data are projected to obtain the column concentration and optical flow labels via a ray marching technique.Gas infrared imaging is simulated based on radiative transfer principles,with data augmentation and background superimposition applied

关 键 词：光学气体成像 速度估计 光流法 合成数据 梯度约束 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]