Measurement of sea-surface velocities from satellite sensor images using the Hopfield neural network

Côté, Stéphane (1996) Measurement of sea-surface velocities from satellite sensor images using the Hopfield neural network. University of Southampton, Doctoral Thesis.

Abstract

The knowledge of ocean surface circulation is of major importance in various scientific applications including the understanding of global meteorological phenomena, resources exploitation, and containment of chemical spills. Ocean surface in situ velocity measurements are usually expensive to obtain. Therefore, the prospect of measuring these velocities from tracking features on sequential satellite sensor images has led to the development of 2 classes of methods, based either on surface feature tracking or on thermal equation inversion.

In this work, a new method for sea-surface feature tracking based on the Hopfield neural network has been developed. The method is based on the minimisation of an energy function that represents the feature tracking problem. It provides several advantages over previous methods, such as the capacity to easily incorporate contextual information and prior knowledge of flow, the flexibility to use images from various data sources, and the capacity to track features of all sizes without compromises. These characteristics are shown to help generate a more physically realistic flow.

Tested on synthetic images with analytical solution, the method was shown to generate better displacement vectors than the popular maximum cross-correlation [MCC] method. Moreover, the method showed a better agreement with in situ velocity measurements than the MCC method for tracking features on real satellite sensor images. The use of contextual information and prior knowledge of flow enables the method to generate smooth and high resolution vector field showing all the details of the velocity field. The new method can be used on various kind of images for tracking, and also find other applications in image registration and pattern recognition. It would therefore be an appropriate candidate for the development of operational systems for the measurement of sea-surface velocities from sequential satellite sensor images.

This record has no associated files available for download.

Contributors

Download statistics