基于单目视觉的石化作业区外因明火感知与测量方法  

Sensing and Measurement Method for External Causes-incurred Open Flames in Petrochemical Operation Area Based on Monocular Vision

作　　者：段志伟[1] 豆全辉 邵女 DUAN Zhi-wei;DOU Quan-hui;SHAO Nv(School of Physics and Electronic Engineering,Northeast Petroleum University)

机构地区：[1]东北石油大学物理与电子工程学院

出　　处：《化工自动化及仪表》2024年第2期199-206,318,共9页Control and Instruments in Chemical Industry

基　　金：国家自然科学基金(批准号:51474069)资助的课题。

摘　　要：针对目前石化作业区外因火灾监测方法大都没有火源定位功能的现状,提出基于单目视觉的外因明火感知与测量方法。首先改进YOLO v7深度学习网络的SiLU激活函数,并引入注意力机制CBAM,使感知网络获得更好的准确性、互适性;随后基于相机成像原理建立物距与像素、分辨率等因子的多元关系测量模型进行距离预测。实验表明:优化后的网络mAP_0.5值提升了2.2%,Precision值提升了5.0%,预测距离误差率绝对值小于3.1%。Considering the fact that most external fire monitoring methods for petrochemical areas have no fire source-positioning function,an external causes-incurred fire sensing and measurement system based on monocular vision was proposed.Firstly,having SiLU activation function of YOLO v7 deep learning network improved and attention mechanism CBAM introduced to raise sensing network's accuracy and interoperability;secondly,having the camera imaging based to establish multivariate relationship measurement model of the object distance,pixel,resolution and other factors so as to predict the distance.Experimental results show that,the m AP_0.5 value and Precision value of the optimized network can be increase by 2.2% and 5.0%,respectively,and the absolute value of predicted distance error rate is less than 3.1%.

关 键 词：单目视觉 明火感知 YOLO v7深度学习 图像处理 注意力机制 多元关系 

分 类 号：TP393.08[自动化与计算机技术—计算机应用技术]