A Hopfield neural network to track drifting buoys in the ocean

A methodology is proposed to estimate sea surface currents from the information given by the sea surface temperature (SST) obtained from satellite images. Currents are estimated from the motion field of a temporal sequence of images using a Hopfield neural network. A cost function is minimized when some rules of correspondence between pixels in successive images are considered. A performance test of their methodology is done using the information from the paths made by Lagrangian buoys. The simulations, done using the buoys positions, prevents the use of a system able to escape from the local minima, where the Hopfield neural network becomes trapped, until reaching a global minima.