留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

基于机器学习方法的三维粒子重构技术

朱浩然 高琪 王洪平 廖相巍 赵亮 魏润杰 王晋军

朱浩然, 高琪, 王洪平, 等. 基于机器学习方法的三维粒子重构技术[J]. 实验流体力学, 2021, 35(3): 88-93. doi: 10.11729/syltlx20200141
引用本文: 朱浩然, 高琪, 王洪平, 等. 基于机器学习方法的三维粒子重构技术[J]. 实验流体力学, 2021, 35(3): 88-93. doi: 10.11729/syltlx20200141
ZHU Haoran, GAO Qi, WANG Hongping, et al. Particle reconstruction of volumetric particle image velocimetry with strategy of machine learning[J]. Journal of Experiments in Fluid Mechanics, 2021, 35(3): 88-93. doi: 10.11729/syltlx20200141
Citation: ZHU Haoran, GAO Qi, WANG Hongping, et al. Particle reconstruction of volumetric particle image velocimetry with strategy of machine learning[J]. Journal of Experiments in Fluid Mechanics, 2021, 35(3): 88-93. doi: 10.11729/syltlx20200141

基于机器学习方法的三维粒子重构技术

doi: 10.11729/syltlx20200141
基金项目: 

国家自然科学基金面上项目 91852204

海洋装备用金属材料及其应用国家重点实验室开放基金课题 SKLMEA-K201910

国家重点研发计划资助 2020YFA0405700

详细信息
    作者简介:

    朱浩然(1990-), 男, 山东嘉祥人, 硕士研究生。研究方向: 基于机器学习的粒子图像测速技术。通信地址: 浙江省杭州市西湖区浙江大学玉泉校区航空航天学院(310027)。E-mail: 21924024@zju.edu.cn

    通讯作者:

    高琪, E-mail: qigao@zju.edu.cn

  • 中图分类号: O357.5+4;TP181

Particle reconstruction of volumetric particle image velocimetry with strategy of machine learning

  • 摘要: 通过三维粒子重构获取粒子场的分布情况是层析粒子图像测速的关键步骤,有限二维投影下的三维粒子重构是一个欠定的反问题,其精确解往往很难得到。一般情况下,可以通过优化方法得到近似解。为了获取质量更高的粒子场并用于层析粒子图像测速,提出了一种基于卷积神经网络(Convolutional Neural Networks,CNN)的粒子重构方法。所提出的技术可以从基于传统的代数重构技术(Algebraic Reconstruction Technique,ART)的方法所得到的粗略粒子分布中进一步提高粒子重构质量。与现有的基于ART的算法相比,新技术在重构质量方面有了显著的改进,可以有效剔除虚假粒子并更准确地还原粒子形状,并且在粒子浓度较稠密的情况下计算速度至少快了一个数量级。
  • 图  1  简单卷积层示意图

    Figure  1.  Sketch of the Simple convolutional layer

    图  2  机器学习方法训练和预测示意图

    Figure  2.  Schematic diagram of machine learning method training and prediction

    图  3  神经网络结构示意图

    Figure  3.  Schematic diagram of neural network structure

    图  4  粒子场截面图

    Figure  4.  Cross-sections of particle field

    图  5  几种方法的质量因子Q变化图

    Figure  5.  Variation of quality factor Q of several methods

  • [1] ELSINGA G E, SCARANO F, WIENEKE B, et al. Tomographic particle image velocimetry[J]. Experiments in Fluids, 2006, 41(6): 933-947. doi: 10.1007/s00348-006-0212-z
    [2] SCARANO F. Tomographic PIV: principles and practice[J]. Measurement Science and Technology, 2013, 24(1): 012001. doi: 10.1088/0957-0233/24/1/012001
    [3] GAO Q, WANG H P, SHEN G X. Review on development of volumetric particle image velocimetry[J]. Chinese Science Bulletin, 2013, 58(36): 4541-4556. doi: 10.1007/s11434-013-6081-y
    [4] DISCETTI S, NATALE A, ASTARITA T. Spatial filtering improved tomographic PIV[J]. Experiments in Fluids, 2013, 54(4): 1-13. doi: 10.1007/s00348-013-1505-7
    [5] WORTH N A, NICKELS T B. Acceleration ofTomo-PIV by estimating the initial volume intensity distribution[J]. Experiments in Fluids, 2008, 45(5): 847-856. doi: 10.1007/s00348-008-0504-6
    [6] ATKINSON C, SORIA J. An efficient simultaneous reconstruction technique for tomographic particle image velocimetry[J]. Experiments in Fluids, 2009, 47(4-5): 553-568. doi: 10.1007/s00348-009-0728-0
    [7] ELSINGA G E, TOKGOZ S. Ghost hunting-an assessment of ghost particle detection and removal methods for tomographic-PIV[J]. Measurement Science and Technology, 2014, 25(8): 084004. doi: 10.1088/0957-0233/25/8/084004
    [8] DE SILVA C M, BAIDYA R, MARUSIC I. EnhancingTomo-PIV reconstruction quality by reducing ghost particles[J]. Measurement Science and Technology, 2013, 24(2): 024010. doi: 10.1088/0957-0233/24/2/024010
    [9] SCHANZ D, SCHRÖDER A, GESEMANN S. 'Shake The Box'-a 4D PTV algorithm: Accurate and ghostless reconstruction of Lagrangian tracks in densely seeded flows[C]//Proc of the 17th International Symposium on Applications of Laser Techniques to Fluid Mechanics. 2014.
    [10] SCHANZ D, GESEMANN S, SCHRÖDER A. Shake-The-Box: Lagrangian particle tracking at high particle image densities[J]. Experiments in Fluids, 2016, 57(5): 1-27. doi: 10.1007/s00348-016-2157-1
    [11] WIENEKE B. Iterative reconstruction of volumetric particle distribution[J]. Measurement Science and Technology, 2013, 24(2): 024008. doi: 10.1088/0957-0233/24/2/024008
    [12] LYNCH K P, SCARANO F. An efficient and accurate approach to MTE-MART for time-resolved tomographic PIV[J]. Experiments in Fluids, 2015, 56(3): 1-16. doi: 10.1007/s00348-015-1934-6
    [13] NOVARA M, BATENBURG K J, SCARANO F. Motion tracking-enhanced MART for tomographic PIV[J]. Measurement Science and Technology, 2010, 21(3): 035401. doi: 10.1088/0957-0233/21/3/035401
    [14] WANG H P, GAO Q, WEI R J, et al. Intensity-enhanced MART for tomographic PIV[J]. Experiments in Fluids, 2016, 57(5): 1-19. doi: 10.1007/s00348-016-2176-y
    [15] GESEMANN S, SCHANZ D, SCHRÖDER A, et al. Recasting Tomo-PIV reconstruction as constrained and L1-regularized nonlinear least squares problem[C]//Proc of the 15th International Symposium on Applications of Laser Techniques to Fluid Mechanics. 2010.
    [16] YE Z J, GAO Q, WANG H P, et al. Dual-basis reconstruction techniques for tomographic PIV[J]. Science China Technological Sciences, 2015, 58(11): 1963-1970. doi: 10.1007/s11431-015-5909-x
    [17] BAJPAYEE A, TECHET A H. Fast volume reconstruction for 3D PIV[J]. Experiments in Fluids, 2017, 58(8): 1-4. doi: 10.1007/s00348-017-2373-3
    [18] BEN SALAH R, ALATA O, TREMBLAIS B, et al. Tomographic reconstruction of 3D objects using marked point process framework[J]. Journal of Mathematical Imaging and Vision, 2018, 60(7): 1132-1149. doi: 10.1007/s10851-018-0800-6
    [19] CAI S Z, ZHOU S C, XU C, et al. Dense motion estimation of particle images via a convolutional neural network[J]. Experiments in Fluids, 2019, 60(4): 1-16. doi: 10.1007/s00348-019-2717-2
    [20] CAI S Z, LIANG J M, GAO Q, et al. Particle image velocimetry based on a deep learning motion estimator[J]. IEEE Transactions on Instrumentation and Measurement, 2020, 69(6): 3538-3554. doi: 10.1109/TIM.2019.2932649
    [21] LAGEMANN C, LAGEMANN K, SCHRÖDER W, et al. Deep artificial neural network architectures in PIV applications[C]//Proc of the 13th International Symposium on Particle Image Velocimetry. 2019.
    [22] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90. doi: 10.1145/3065386
    [23] CHEN L C, PAPANDREOU G, KOKKINOS I, et al. DeepLab: semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(4): 834-848. doi: 10.1109/TPAMI.2017.2699184
    [24] QIAN N. On the momentum term in gradient descent learning algorithms[J]. Neural Networks, 1999, 12(1): 145-151. doi: 10.1016/S0893-6080(98)00116-6
    [25] TIELEMAN T, HINTON G. Lecture 6.5-rmsprop: divide the gradient by a running average of its recent magnitude[R]. COURSERA: Neural Netw Mach Learn, 2012, 4(2): 26-31.
    [26] KINGMA D, BA J. Adam: a method for stochastic optimization[C]//Proc of the 3rd International Conference on Learning Representations (ICLR). 2015.
    [27] IOFFE S, SZEGEDY C. Batch normalization: Accelerating deep network training by reducing internal covariate shift[J]. Computer Science, 2015, arXiv: 1502.03167. http://dl.acm.org/citation.cfm?id=3045118.3045167
    [28] ABADI M, BARHAM P, CHEN J, et al. Tensorflow: A system for large-scale machine learning[C]//Proc of the 12th USENIX Symposium on Operating Systems Design and Implementation (OSDI 16). 2016.
  • 加载中
图(5)
计量
  • 文章访问数:  798
  • HTML全文浏览量:  251
  • PDF下载量:  115
  • 被引次数: 0
出版历程
  • 收稿日期:  2020-11-10
  • 修回日期:  2021-01-10
  • 刊出日期:  2021-06-25

目录

    /

    返回文章
    返回

    重要公告

    www.syltlx.com是《实验流体力学》期刊唯一官方网站,其他皆为仿冒。请注意识别。

    《实验流体力学》期刊不收取任何费用。如有组织或个人以我刊名义向作者、读者收取费用,皆为假冒。

    相关真实信息均印刷于《实验流体力学》纸刊。如有任何疑问,请先行致电编辑部咨询并确认,以避免损失。编辑部电话0816-2463376,2463374,2463373。

    请广大读者、作者相互转告,广为宣传!

    感谢大家对《实验流体力学》的支持与厚爱,欢迎继续关注我刊!


    《实验流体力学》编辑部

    2021年8月13日