Wake Detection Using Convolutional Neural Networks
Abstract
Advances in engineering and technology have made acoustic detection of submarines increasingly difficult. Using hydrodynamic signatures created by propagating submarines is an alternative method for submarine detection. Detection based on hydrodynamic signatures may also offer unique tactical advantages, given the tendency of wakes to persist for long timescales. Artificial neural networks trained on velocity field data show promise to automate detection. We used numerical simulations to generate velocity data on wake, jet, and convective turbulence. All these forms of turbulence have similar characteristics in the velocity field, yet their spectra reveal subtle differences that could be exploited for wake identification purposes. We then trained a convolutional neural network with the simulation results and demonstrated that neural networks can classify turbulent flows based on small-scale features with high accuracy. In particular, we find that the developed algorithms can successfully identify wakes in 92% of cases, which implies that the AI-based technology is viable and ready for the transition to the analysis of the experimental and field data.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 2021
- Accession Number
- AD1165039
Entities
People
- Jacqueline Zimny
Organizations
- Naval Postgraduate School