A Deep Learning Based Denoising Approach for Synthetic Aperture Sonar Imaging

Synthetic Aperture Sonar (SAS) is one of the techniques used for underwater imaging, and in this paper, we use the range Doppler algorithm for SAS imaging and consider the position errors of sonar platforms in practical scenarios. These errors can seriously degrade the imaging result; however, correcting them and performing compensation with high-precision equipment could be very costly. In this paper, a deep learning approach is proposed using artificial neural networks for image denoising, specifically targeting noise induced by motion uncertainties. As demonstrated by computer simulations, it can recover the images with better quality without employing complex algorithms and hardware compensation.