Estimation of ocean surface currents from satellite imagery using a Hopfield neural network

A new approach for estimating sea-surface currents from sequential satellite sensor images has been developed. Thermal gradients from two sequential images were enhanced and mathematically described as a set of features. A Hopfield neural network was used to match the features through the minimization of an energy function defining the matching problem. The method was tested on tracking sea-surface thermal gradients and compared to the Maximum Cross-Correlation (MCC) method. It appears that the neural network method is more appropriate for measuring currents in areas presenting high local variations of current direction. and intensity. Moreover, it shows potential for direct calculation of cross- and along-isothermal components of the currents, for obtaining velocity of rotating features and of deforming features that are difficult to obtain with the MCC method. It can also easily be adapted for tracking other features such as ice floes or clouds. The new method would therefore be more appropriate than the MCC method for development of automatic and reliable operational systems for the measurement of sea-surface velocities from sequential satellite sensor images.