Platform motion estimation in multi-band synthetic aperture sonar with coupled variational autoencoders.

Area of Science:

• Underwater acoustics
• Signal processing
• Machine learning

Background:

• Coherent processing in synthetic aperture sonar (SAS) necessitates highly accurate platform motion estimation and compensation for high-resolution imaging.
• Micronavigation, a through-the-sensor estimation method, is critical when external positioning data is unreliable or unavailable.

Purpose of the Study:

• To develop an advanced unsupervised machine learning method for micronavigation in multi-band SAS systems.
• To enhance the accuracy of platform motion estimation by leveraging the multiple-input multiple-output (MIMO) configuration.

Main Methods:

• Utilized a machine learning approach based on variational Bayesian inference for unsupervised, data-driven micronavigation.
• Employed a hierarchical variational inference scheme combined with the MIMO arrangement of a multi-band SAS system.
• Implemented a self-supervision mechanism within the inference scheme to learn platform motion.

Main Results:

• Achieved improved micronavigation accuracy through the proposed hierarchical variational inference method.
• Demonstrated the effectiveness of exploiting the MIMO configuration for enhanced motion estimation.
• Validated the unsupervised, data-driven nature of the learning process.

Conclusions:

• The developed hierarchical variational inference method significantly enhances micronavigation accuracy in multi-band SAS.
• Exploiting MIMO configurations within a self-supervised learning framework offers a robust solution for precise platform motion estimation.
• This approach is particularly valuable for high-resolution SAS imaging in environments with limited or inaccurate external navigation data.