基于图像动态纹理特征的气固流化床流型识别  

Flow regime identification of gas-solid fluidized bed based on images dynamic texture characteristics

作　　者：周云龙[1] 李莹[1] 赵红梅[1] 

机构地区：[1]东北电力大学能源与动力工程学院,吉林吉林132012

出　　处：《化学工程》2011年第12期59-63,共5页Chemical Engineering(China)

摘　　要：准确识别流型是气固流化床二相流参数检测的重要内容,文中提出一种基于图像光流法和动态纹理特征相结合的气固流化床流型识别的新方法。实验是在气固流化床二相流实验系统上利用高速摄影系统获取流型图像。流型图像分别为鼓泡床,节涌床,湍动床,快速流化床,稀相输送等5种典型流型。首先对获取的不同流型图像分别进行去噪和对比度拉伸等预处理,然后运用光流法得到连续2帧图像的光流场,再通过灰度共生矩阵提取图像的动态纹理特征,作为流型识别的输入特征向量。并分别结合弹性BP神经网络,Elman神经网络,BP神经网络进行训练,实现流型的识别。实验表明:动态纹理特征和弹性BP网络相结合的方法更能有效地识别气固流化床中的5种典型流型,整体识别率达到98%,为流型识别开辟了一种新方法。The exact identification of flow regime is an important content to detect the parameters of gas-solid two- phase flow fluidized bed. A new flow regime identification method based on images optical flow technique and dynamic texture was proposed. The experiment was conducted on gas-solid fluidized bed system and the flow images were captured by a high speed photography system. The flow images are of five typical regimes of gas-solid twophase flow fluidized bed, including bubbling bed, slugging bed, turbulent bed, wall pressing flow, and thin phase conveying. First, the different flow images captured were optical flow technique was used to get the optical flow field pretreated by denoising and contrast stretching, then, of continuous two frames images. The image dynamic texture characteristics were extracted by gray level co-occurrence matrix, regarded as input variable. Those samples were separately sent to elasticity BP neural net, Elman neural net and BP neural net work for optimization. Thus the image texture eigenvectors of flow regime were identified. The experimental results show that the combination between dynamic textures and elasticity BP neural net can more effectively identify the five typical regimes of gassolid two-phase flow fluidized bed. The whole identification accuracy is 98%, opening up a new avenue of flow pattern recognition.

关 键 词：气固流化床 流型识别 光流法 灰度共生矩阵 图像处理 

分 类 号：O359[理学—流体力学]