ART2 networks for particle image velocimetry

Artificial neural networks (ANNs) are novel computing architectures which are increasingly being applied to particle image velocimetry. The choice of a suitable range of features to represent particles for recognition can be an intractable problem. The effectiveness of size, position, shape and intensity measurements as suitable representations of particles when using the ART2 network for particle tracking is studied. A novel technique that uses two ART2 networks for particle tracking and error suppression, is presented. The processed images used were those of liquid crystal particles in a natural convective flow taken at successive time intervals. This method has been successfully applied to the determination of displacements in simulations of uniform flows. The fidelity of the network in tracking particles generally decreases with increasing displacements.