Hopfield neural network PTV with new strategy
Kazuo Ohmi, Achyut Sapkota · Journal of the Visualization Society of Japan · 2004
The Hopfield neural network was already applied with success to the particle pairing process by the present authors. However it was limited to the frame pairs with relatively less number of particles due to its high computation time. In the present study the algorithm has been improved for reduced computation time. The point of the improved algorithm is new concept in connection scheme of the neurons, according to which the network is reduced from fully connected to partially connected scheme. The constrained conditions for the object function are further modified in order to meet with this new strategy. As a result the algorithm is able to handle the higher number of particles with higher degree of particle pairing accuracy within much shorter computation time. The algorithm is tested by using the PIV Standard Image of the Visualization Society of Japan,