Adaptive Sampling for Seabed Identification from Ambient Acoustic Noise

Le résumé fourni par la source

This thesis discusses distinguishing geoacoustic properties of seabeds by adaptively sampling ambient acoustic measurements using an autonomous underwater vehicle. Modern advancements in seabed characterization motivate an online method for distinguishing between two seabed types using Fourier transformed ambient acoustic noise snapshots. These snapshots are assumed to be associated with their spatial location and seabed, making the goal to organize unlabeled seabed locations. The snapshots are transformed to obtain pairwise similarities between locations. Locations with similarities exceeding a threshold are classified together with the goal of identifying all locations with similarities below this threshold using a process known as level set estimation. We propose an adaptive sampling policy that aims to directly reduce the number of locations with unknown level set membership via a lookahead step in addition to minimizing the distance traveled through a nearest neighbors approach. Results on synthetic, single-boundary fields and multi-boundary realistic world data are evaluated by comparing predicted and true level set assignments and demonstrate the benefits of the proposed algorithm. Furthermore, extensions discussing the effects of a two-step and path planning lookahead in addition to the benefits of bathymetry data are included to motivate further research.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.