Hopfield Neural Network for Sea Surface Current Tracking from Tiungsat-1 Data

This paper introduces a new approach for neural network application to coastal studies. The method is based on the utilization of the Hopfield neural network to model sea surface current movements from single TiungSAT-1 image. In matching process using Hopfield neural network, identified features have to be mathematically compared to each other in order to build an energy function that will be minimized. In this context, the neuron network has been taken in two dimensions; raw and column in order to match between the similar features of surface pattern. It was required that the two features were extracted from the same location. The Euler method is used to minimized the energy function of neuron equation. The study shows that the surface current features such as structure morphology of water plume can be automatically detected. In TiungSAT-1 data, green and near-infrared bands were competent at sea surface current features detection with high accuracy speed of ±0.14 m/s. It can be said that, Hopfield neural network has highly promised feature enhancement and detection in optical satellite sensor such as TiungSAT-1 image. In conclusion, Hopfield neural network can be used advance computational tool for modeling the pattern movement of sea surface in satellite data.