Logo image
Back
Enhancing Underwater Sonar Target Recognition using Non-linear Wavelet Transforms
Conference proceeding

Enhancing Underwater Sonar Target Recognition using Non-linear Wavelet Transforms

Andrew Christensen, Ananya Sen Gupta and Ivars Kirsteins
International Symposium on Ocean Electronics (Print), pp.1-6
12/10/2025
DOI: 10.1109/SYMPOL68153.2025.11395927

View Online

Abstract

This work addresses a critical challenge of automatic target recognition in sonar by focusing on feature extraction and feature selection techniques. The inherent non-linear distortions prevalent in sonar echo returns, arising from complex acoustic scattering physics, waveguide propagation effects, and background clutter, severely limit the effectiveness of standard off-the-shelf machine learning classifiers applied to raw echo data. To overcome these limitations and extract highly informative target features, we propose using a non-linear feature extraction method to first separate the target signatures from the background distortions. Specifically, this method utilizes a cascade of wavelet filters combined with non-linear operators, yielding a set of features more amenable to linear classification. This non-linear wavelet transform is well-suited for sonar echo returns due to its highly localized time-frequency characteristics. Our method achieved a classification accuracy of 86.6% on a large sonar target dataset, representing a 13.9% improvement in performance compared to using only the raw sonar echoes. These results demonstrate the discriminatory power of the wavelet-based feature extraction techniques for challenging sonar ATR applications.
 Machine Learning Physics Acoustic distortion automatic target recognition Feature extraction Nonlinear distortion sonar Sonar applications Target recognition Time-frequency analysis Wavelet transforms wavelets

Details

Metrics

1 Record Views
Logo image