Inferring depth contours from sidescan sonar using convolutional neural nets

Abstract

Sidescan sonar images are 2D representations of the seabed. The pixel location encodes distance from the sonar and along track coordinate. Thus one dimension is lacking for generating bathymetric maps from sidescan. The intensities of the return signals do, however, contain some information about this missing dimension. Just as shading gives clues to depth in camera images, these intensities can be used to estimate bathymetric profiles. The authors investigate the feasibility of using data driven methods to do this estimation. They include quantitative evaluations of two pixel-to-pixel convolutional neural networks trained as standard regression networks and using conditional generative adversarial network loss functions. Some interesting conclusions are presented as to when to use each training method.

Neural Shape-from-Shading for Survey-Scale Self-Consistent Bathymetry from Sidescan

“…Before discussing the problem of bathymetry estimation from sidescan, it will be instructive to first consider the inverse problem of simulating sidescan. As it is often simpler to reconstruct sidescan from bathymetry, this is usually the way in which the former problem is phrased, with one notable exception being data-driven models [6]. Model-driven approaches [7] [8] [9] [10] [11] take as input seafloor geometry, possibly together with some relevant acoustic properties.…”

Efficient imaging method for multireceiver SAS

“…The second term is introduced by the error of the 'stop-and-hop' approximation. In general, the displaced distance between the transmitter and receiver does not affect the phase shown in (4). At this point, this term is called the QM term.…”

“…Synthetic aperture sonar (SAS) [1–3] dramatically improves the azimuth resolution. Due to this reason, this technique can be used for the seabed mapping [4, 5], marine archaeological surveying [6], target classification [7, 8], and so on. However, the traditional monostatic SAS system configured with a single transmitter and a single receiver possesses the contradiction between the wide swath coverage and high resolution in the azimuth dimension.…”

Bathymetric Reconstruction From Sidescan Sonar With Deep Neural Networks

“…The sidescan intensities do contain some information about the seafloor's material and elevation changes: harder materials tend to have higher return intensities; the nadir range in every ping gives a single altitude reading; and, more importantly, the intensity changes indicate the change of the incidence angle [1]. Even though the unknown bottom material and other disturbances make it difficult to estimate the depth from sidescan returns analytically, data-driven methods [2] have shown promising results in estimating depth contours from sidescan intensities.…”