Acoustic identification of buried underwater unexploded ordnance using a numerically trained classifier (L)

Area of Science:

• Underwater acoustics
• Machine learning applications
• Geophysical surveying

Background:

• Unexploded ordnance (UXO) poses significant risks in marine and coastal environments.
• Accurate detection of buried UXO, such as rockets, is crucial for safe clearance operations.
• Acoustic scattering methods offer a non-invasive approach to subsurface object identification.

Purpose of the Study:

• To develop and validate a machine learning algorithm for identifying buried unexploded ordnance rockets using acoustic scattering data.
• To simulate acoustic scattering from rockets at various burial angles to create a comprehensive training dataset.
• To test the algorithm's performance in a controlled sediment environment with diverse targets.

Main Methods:

• Finite element-based structural acoustics code was employed for acoustic scattering simulations.
• Simulations covered 90 rocket burial angles at 2° increments.
• A generative relevance vector machine (RVM) algorithm was trained using simulated data and tested with experimental measurements.

Main Results:

• The trained RVM algorithm demonstrated successful identification of buried rockets.
• The algorithm was validated against experimental scattering measurements in a sediment pool.
• Performance was assessed for various targets, including rockets at different angles, a boulder, and cinderblocks.

Conclusions:

• Machine learning, specifically RVM, can effectively identify buried unexploded ordnance rockets from acoustic scattering data.
• Simulated acoustic scattering data is valuable for training robust detection algorithms.
• The developed method shows promise for real-world UXO detection in underwater environments.