Neural Network for Surface Current Trajectory
Modeling From RADARSAT-1 SAR Data
Maged Marghany
Laboratory of Physical Oceanography
Institute of Oceanography (INOS)
Universiti Kolej Sains Dan Teknologi Malaysia (KUSTEM)
21030 Mengabang Telipot, Kuala Terengganu,
Malaysia
Email: mmm@kustem.edu.my,
magedupm@hotmail.com
Abstract
This paper introduces a new approach for utilizing the neural network for real time surface
current simulation from RADARSAT-1 SAR image. The surface current parameters are
collected by using AWAC during the RADARSAT-1 SAR pass over. The neural network input
is a vector containing the values of the RADARSAT-1 SAR image intensity gradients. In this
paper, a single feed forward -propagation neural network was utilized to estimate the Doppler
frequency shift in order to determine the surface current pattern along RADARSAT-1 SAR
image. It is found that, the neural network outperformed conventional regression technique in
modeling surface current velocity and their directions. The RMS detected from NN model was
0.02 m/s. The reduction of the amount of the errors is due to good performance of regression
model.
1.0 Introduction
Recently, scientists and researchers have paid a great attention in utilization of NN in modeling
environmental problems. An artificial neural network (NN) can be identified as a mathematical
model consists of many non-linear computational elements, named neurons. These neurons are
operating in parallel and massively connected by links characterized by different weights. A
single neuron computes the sum of its inputs, adds a bias term, and drives the result through a
generally nonlinear activation function to produce a single output termed the activation level of
that neuron. NN models are essentially specified by net topology neuron distinctiveness, and
instruction or information system (Lim et al., 2000).
A key question is how NN cab be used to simulate the current trajectory movements from a
single SAR image. The main objective of this study is to exploit the NN algorithm to simulate
the surface current movements; in this study, NN based simulation of surface current pattern is
demonstrated using the RADARSAT-1 image which simulated from multi-data. The developed
NN algorithm is used to obtain spatial distribution of the surface current movements.
