Artificial neural network-based inversion technique for extracting ocean surface wave spectra from SAR images

Abstract

An artificial neural network (ANN) based nonlinear technique for inverting the SAR image spectrum of ocean surface waves is developed. In this technique, a multi-layer perceptron (MLP) is used to perform the inversion process. The MLP is trained using simulated SAR and wave spectra. The training process utilizes the standard error-backpropagation technique. The results indicate that the method works well over a large range of wind and wave conditions. The error in the inversion process was found to increase in the higher sea states. The technique works best if the network is used within the range over which it was trained. It is noted that this technique may be used independent of SAR imaging models, by training the network with coincident and co-located measurements of SAR and wave spectra.