A study of prediction in seabed mapping

The aim of this research is to investigate the use of the modern prediction algorithms in seabed mapping. These prediction algorithms can be used in enhancing the quality of the measured bathymetric data to help in filtering the measured data and excluding noise  from  actual  data  to  ensure  higher  seabed  mapping  accuracy  for  more  secure navigation. The work involves the development of a general purpose sonar simulation platform to generate   the   required   data   for   testing   the   different   prediction   algorithms.   The simulation  platform  consists  of  a  seabed  simulator and  an  interferometric  sonar simulator for bathymetric measurements. Two methods of building the seabed simulator have been investigated and applied; the fractal   geometry   based   method   and   the   random   generator   based   method.   The interferometric sonar simulator is based on SAS (Synthetic Aperture Sonar) which is a widely accepted modern technology. The  predictors  investigated  in  this  work  are  based on  KNN  (K  Nearest  Neighbours) and  dynamic  ANN  (Artificial  Neural  Networks).  Both dynamic  feedforward  and dynamic   recurrent   networks   are   investigated.   The   comparison   between   the performances  of  these  different  predictors  reveals that  dynamic  recurrent  networks outperform   the   other   types   of   predictors   and   the   Nonlinear   AutoRegressive eXogenous (NARX) Neural Network is the best