A Hopfield neural network approach for the reconstruction of wide-bandwidth sonar data

Sonar systems with small physical apertures are easier to mount on small vessels and remotely operated vehicles (ROVs). Such systems however are limited in terms of angular resolution. Although wide-bandwidth signals may be used to increase the range resolution of a sonar system, angular resolution is unaffected. Such limitations can be overcome if the region of interest in the underwater environment is insonified from a number of different angles, and this low resolution information reconstructed into a high resolution image of the region. This paper proposes a reconstruction approach based on a Hopfield neural network. This approach is shown to perform better than the inverse Radon transform for image reconstruction under both noisy and noise-less conditions. To verify these claims, results are presented using both real and simulated sonar data.