A new neural network for particle-tracking velocimetry
A new neural network for particle-tracking velocimetry
复制标题
DOI：
10.1007/s003480050297
复制
发表时间：
1999-03-01
影响因子：
2.4
通讯作者：
Labonté, G
中科院分区：
文献类型：
--
作者：
Labonté, G
我们描述了一种新的神经网络设计来解决粒子跟踪测速的对应问题。给定两张连续的悬浮在流体中的标记粒子的图片，它通过近似复制流体运动来匹配它们的图像。我们提出了效率测试的结果，揭示了其性能的卓越性和其稳定性，相对于存在的不可匹配的粒子图像。我们将其在图像匹配中的成功率与Grant和Pan（1995）的神经网络进行了比较，并观察到当流动方向发生更重要的变化时，它会产生更好的结果。它具有优于后者的重要优点，即更好地适应于受益于并行计算，并且是自启动的，即不需要预先教导流体流动。
We describe a new neural network designed to solve the correspondence problem of particle-tracking velocimetry. Given two successive pictures of marker-particles suspended in a fluid, it matches their images by approximately duplicating the fluid motion. We present the results of efficiency tests that reveal the excellence of its performance and its stability with respect to the presence of unmatchable particle images. We compare its success rate in image matching to that of the neural network of Grant and Pan (1995), and observe that it produces better results when the flows have more important changes in direction. It has the important advantages over the latter, of being better adapted to benefit from parallel computing, and of being self-starting, i.e. of not requiring to be taught about the fluid flow in advance.