250-kHz GeoSwath 4 data comparisons were done for an Alpine lake, from the shoreline out to 120-m depth. This across-track bathymetric profile shows that it is possible to achieve a “clean” 350-m swath width in 80 m of water. sulting in a wreck being located at 30-m depth in the data map. The wreck can be seen with both manual and AI filtering applied. While subtle differences are possible to spot, the images are essentially identical, with both pro-viding high quality and actionable data to the required standard. This demonstrates that AI filtering matches the results of manual filtering. More 250-kHz GeoSwath 4 data comparisons were done in a completely different environment: on an Al-pine lake in October 2021, from the shoreline out to 120-m depth. An across-track bathymetric profile from the survey shows that it is possible to achieve a “clean” 350-m swath width in 80 m of water. For this application, the results have been described as “a game changer” by the system owner and data consumer. Real-Time Performance Now that the AI is proven to be able to provide a sounding-acceptance solution in a repeatable and us-er-independent manner, it is also possible to calculate the total vertical uncertainty associated with the sonar data, which could be fed directly into TPU-and cube-based data processing workflows. Together with improved real-time outlier identifica-tion, sonar-dependent IHO conformance can also be displayed in real time, which allows users to check that data collection likely meets the required spec and adjust the survey on the fly, which saves both time and money. The new AI data processing module for the GeoSwath 4 bathymetric sonar is clearly meeting the goal to opti-mize the acquisition and availability of precise seabed data. The AI project has delivered amazing results that can be repeated on any relevant data set without issue. A new generation of hands-free filters for GeoSwath 4 data processing will be the product of commercialization of this joint research project. The work is almost complete, and GeoSwath 4 owners will be able to integrate AI pro-cessing into their systems without charge later this year. References For a list of references, contact Francisco J. Gutiérrez at: francisco.gutierrez@geoacoustics.com. ST Francisco J. Gutiérrez is a product specialist for the GeoSwath 4, Pulsar and TOPAS systems manufactured by GeoAcoustics Ltd. He previ-ously worked as technologist for the Spanish Re-search Council and as an associate professor of electronics at University of Cadiz in Spain. Danny Websdale is an AI/machine learning re-search scientist at the University of East Anglia (UEA) and GeoAcoustics Ltd. He received a bachelor’s degree in computer systems engi-neering (2014) and a Ph.D. in audio-visual speech processing (2019), both from UEA. He is currently working on applying AI for real-time seabed classification. www.sea-technology.com June 2022 | ST 13
Sea Technology June 2022: Page 13
