Related Experiment Video
Updated: Aug 6, 2026
04:54
A Stable Phantom Material for Optical and Acoustic Imaging
Published on: June 16, 2023
Knowledge-inspired and sample-generation-based spectral augmentation for few-shot underwater acoustic target
Wei Gao1, Desheng Chen1, Junhui Zhang1
1School of Information and Electronics, Beijing Institute of Technology, Beijing 100081, China.
The Journal of the Acoustical Society of America
|July 16, 2026
Summary
This study introduces a novel framework for underwater acoustic target recognition, enhancing few-shot learning with advanced data augmentation and recalibration techniques to improve accuracy in noisy, data-scarce conditions.
Area of Science:
- Marine acoustics
- Machine learning
- Signal processing
Background:
- Underwater acoustic target recognition faces challenges from data scarcity and non-stationary noise.
- Existing few-shot learning methods struggle with low signal-to-noise ratios and complex marine environments.
- Effective exploitation of spectrotemporal characteristics is crucial for robust recognition.
Purpose of the Study:
- To propose a sample-generation-based few-shot learning framework for underwater acoustic target recognition.
- To enhance recognition accuracy and robustness, particularly under low signal-to-noise ratio and cross-domain conditions.
- To address limitations in current few-shot methods for exploiting structured spectrotemporal features.
Main Methods:
- A knowledge-inspired spectral decomposition augmentation strategy creates multi-view samples in the spatial-frequency domain.
- A Siamese network with time-frequency attention learns robust spectrotemporal representations.
- A cross-feature guided recalibration module dynamically refines augmented views using original features.
- Multi-view joint learning and task-adaptive fine-tuning with pseudo-sample generation are employed.
Main Results:
- The proposed method significantly improves recognition accuracy and robustness on the ShipsEar and DanShip datasets.
- Superior performance is demonstrated under low signal-to-noise ratio and cross-domain conditions compared to state-of-the-art methods.
- The framework effectively suppresses noise-sensitive variations while preserving essential spectrotemporal structures.
Conclusions:
- The developed framework offers a powerful solution for few-shot underwater acoustic target recognition.
- The spectral decomposition augmentation and cross-feature recalibration modules are key to achieving enhanced performance.
- The method shows strong potential for real-world applications in complex marine environments.
