基于RPCA及低秩表示的气液两相流动图像中气泡图像分离研究  

Flow Image Based on Low-Rank Approximations

作　　者：魏文君 李海广[1] 郭旭凯 吴晅[1] WEI Wenjun;LI Haiguang;GUO Xukai;WU Xuan(School of Energy and Environment,Inner Mongolia University of Science and Technology,Baotou 014010,China)

机构地区：[1]内蒙古科技大学能源与环境学院,包头014010

出　　处：《真空科学与技术学报》2023年第4期339-349,共11页Chinese Journal of Vacuum Science and Technology

基　　金：国家自然科学基金项目(51666015)。

摘　　要：气液两相流中对气泡的测量研究是非常重要的,气泡测量技术中,如何实现气泡与背景分离是研究的重点问题。现有的测量技术大多采用图像二值化、边缘检测、图像滤波等方法来实现气泡信息的提取,而这些测量方法往往是存在不足的,仅仅针对单一图片或者需要人为手动选取。本文通过SVD(单值分解)和RPCA(鲁棒主成分分析法)对气液两相流中的气泡图像进行背景分离,其方法主要有两个特点:连续相关性和自动获取性。并提出逐行累加和逐列累加的方法,测量气泡的运动过程形态。研究表明,相比于原始的图像分离技术,利用RPCA运算,对气泡的定位、大小和速度表示都更准确。As we all know,two-phase flow widely exists in modern industrial processes and everyday life.Interaction between the gas phase and fluid phase exists in the gas-fluid flow,and its complex fluid flow characteristics make it difficult to detect the two-phase flow parameter.In the process of traditional flow image calculation,we can not get accurate results because of the complex background.Therefore,an improved algorithm based on singular value decomposition and robust principal component analysis(RPCA)is proposed and applied to saliency analysis and feature extraction of flowing images.This algorithm has two features:feature extraction and anti-jamming.Experimental results show that the proposed algorithm has better detection performance than existing flow image detection algorithms and has lower time complexity.

关 键 词：气液两相流 SVD分解 RPCA主成分分析 特征提取 

分 类 号：O35[理学—流体力学]