基于图像轮廓特征和粒子群优化神经网络的气液两相流流型识别  

Identification method of gas-liquid two-phase flow pattern based on image contour feature and particle swarm optimized neural network

作　　者：周云龙[1] 李洪伟[2] 陈飞[2] 

机构地区：[1]东北电力大学能源与机械工程学院,吉林省吉林市132012 [2]东北电力大学自动化工程学院,吉林省吉林市132012

出　　处：《吉林大学学报（工学版）》2009年第3期673-678,共6页Journal of Jilin University:Engineering and Technology Edition

基　　金：吉林省科技发展计划项目(20040513)

摘　　要：根据流型轮廓特征提取的不变矩能有效检测出具有平移、旋转、比例变化的图像特性,提出了一种基于图像不变矩与粒子群优化神经网络相结合的气液两相流流型识别的新方法。该方法利用高速摄影系统获取水平管道内气液两相流的流动图像,经过图像处理后和形态学轮廓提取图像不变矩特征向量,并以此特征向量作为流型样本对粒子群优化神经网络进行训练,实现了对流动图像的流型智能化识别。实验结果表明,训练成功的粒子群优化神经网络能够快速准确识别水平管道内的7种典型流型,整体识别率达到97.5%,为流型在线识别提供了一种新的有效方法。According to the fact that the moment invariant extracted from the flow pattern image contour can identify the image characteristies with translation,rotation and scale variation,a flow pattern identification method was proposed based on the image moment invariant and the particle swarm optimized neural network.The gas-liquid two-phase flow pattern images in a horizontal pipe were captured by a high-speed digital video system.The image moment invariant eigenvectors were extracted using the image processing ...

关 键 词：液体力学 流型识别 图像处理 轮廓提取 仿射不变矩 粒子群优化神经网络 

分 类 号：O359.1[理学—流体力学]