Autonomous Optical Survey Based on Unsupervised Segmentation of Acoustic Backscatter

Abstract

The application of acoustics to study the seabed have for decades provided industry and science with valuable information, and is still excels in terms of spatial coverage and detail. An acoustic response from the seabed not only contains information about the range, through the two way travel time, but also the acoustic reflectivity of the substrate from the strength of the backscatter response. As the signal strength differs between substrate types, this information can be used to detect and classify different seabed types. However, there are ambiguities in the acoustic signatures and the reliance on ground truth samples, for succeeding in this identification, is a limiting factor. In this paper we present a way to mitigate this problem using Hidden Markov Random Fields (HMRF) to perform unsupervised segmentation of the backscatter response for the purpose of determining different seabed types. The outcome of this analysis is directly used to plan and conduct an autonomous near-seabed camera survey to verify the classification results, whilst complementing the acoustical data-set. The method is tested in a full-scale experiment and performed in-situ onboard a Kongsberg Hugin 1000 autonomous underwater vehicle (AUV).