The Hopfield neural network as a tool for feature tracking and recognition from satellite sensor images

Abstract. A new approach for feature tracking and recognition on sequential satellite sensor images using neural networks has been developed. Feature tracking is recognized as being of importance in applications such as ice-mapping, cloud motion winds, ocean currents, and short-term forecasting. Feature recognition finds application in automatic image navigation. This paper explores the potential of a Hopfield neural network to perform feature tracking or recognition, and gives examples of its implementation to three different applications. It is shown that the net can provide superior performance to existing techniques for tracking, the advantages of this new approach being its precision, speed, low sensitivity to deformation, its capacity to detect rotational motion, and to provide directly the cross- and along-isopycnal components of displacement vectors. It is also shown that the Hopfield neural network can provide a valuable tool for automatic image navigation through coastline recognition.