基于卷积神经网络的水中大尺寸气泡体积测定  

Volume measurement of large-sized bubbles in water based on convolutional neural network

作　　者：郝宗睿 李臣豪 索铜声 陈杰 华志励 HAO Zong-rui;LI Chen-hao;SUO Tong-sheng;CHEN Jie;HUA Zhi-li(Institute of Oceanographic Instrumentation,Qilu University of Technology(Shandong Academy of Sciences),Qingdao 266100,China)

机构地区：[1]齐鲁工业大学(山东省科学院)海洋仪器仪表研究所,山东青岛266100

出　　处：《齐鲁工业大学学报》2022年第5期10-15,50,共7页Journal of Qilu University of Technology

基　　金：山东省重点研发计划项目(2019GHY112064)。

摘　　要：由于大尺寸气泡在静止流场内受粘滞力和表面张力的作用,呈现出不规则的几何形状,导致按照常规计算方法推算出的气泡等效体积存在误差。为了降低大气泡体积计算的误差,通过高速相机捕获静止流场内的气泡,将经图像预处理后的气泡图像作为改进型卷积神经网络模型的输入,利用小气泡的二维投影面积和对应三维体积作为数据集对改进模型进行网络训练,训练后模型识别准确率高达98.87%。用训练后的网络模型预测不规则大尺寸气泡的三维体积,体积预测的准确率达94.76%。Due to the effect of viscous force and surface tension in the static flow field,the large-sized bubbles present an irregular geometric shape,which leads to errors in the equivalent volume of the bubbles calculated by the conventional calculation method.In order to reduce the error in the calculation of the volume of large bubbles,the bubbles were captured in the static flow field with a high-speed camera system in this work.The bubble image after image pretreatment was used as the input of the improved convolutional neural network model.Using the two-dimensional projected area and the corresponding three-dimensional volume as a data set,the improved model was trained on the network,and the recognition accuracy of the model after training was up to 98.87%.The three-dimensional volume of irregular large-sized bubbles was predicted by the trained network model,and the accuracy rate of volume prediction was 94.76%.

关 键 词：卷积神经网络(CNN) 气泡体积 图像处理 模型训练 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]