AI compresses AUV seafloor images for low-bandwidth satellite transmission
A 400,000x data reduction lets 2.8 hours of imagery transmit in 34 minutes.
A team led by Adrian Bodenmann (University of Southampton) and co-authors from the National Oceanography Centre, NATO STO-CMRE, and others published a method for remote awareness of seafloor images collected by AUVs over low-bandwidth communication links. The technique uses AI to either select a set of images that best represent an entire dataset or find images most similar to a given query. These selected images, combined with metadata from the full set, are then compressed and transmitted via satellite or underwater acoustic modems, giving shore operators real-time insight into what the AUV is collecting while it's still deployed.
Field validation came from three deployments off the UK coast and in Gran Canaria using different AUVs and imaging systems. The method achieved an almost 400,000-fold reduction in data volume compared to raw data. For a 2-hour-47-minute mapping mission, the data summary transmitted in just over 34 minutes over low-bandwidth satellite communication. This dramatic compression enables operators to adapt mission plans on the fly based on transmitted imagery, improving efficiency and reducing the need for costly recovery and re-deployment. The work bridges machine learning, robotics, and image processing for practical ocean exploration.
- AI selects representative or query-matching seafloor images from AUV datasets in real time
- Achieves 400,000-fold data reduction over raw image sizes
- Transmits a 2-hour-47-minute mission summary in 34 minutes via satellite
Why It Matters
Enables real-time remote awareness for autonomous ocean exploration, drastically cutting transmission costs and latency.