Related Experiment Video
Updated: Jun 11, 2026
06:42
Conducting Hyperscanning Experiments with Functional Near-Infrared Spectroscopy
Published on: January 19, 2019
High-SNR phase unwrap method based on RMN-LSTM for distributed hydro-acoustic detection
Applied Optics
|June 10, 2026
Summary
A novel RMN-LSTM neural network enhances underwater target detection by accurately reconstructing ship acoustic fingerprints from noisy data. This method significantly improves signal quality without requiring hardware upgrades.
Area of Science:
- Marine acoustics
- Signal processing
- Artificial intelligence
Background:
- Distributed acoustic sensing (DAS) is promising for underwater target detection.
- Ship detection relies on line-spectrum components, which are vulnerable to marine noise.
- Noise degrades the signal-to-noise ratio (SNR), causing phase distortions in DAS data.
Purpose of the Study:
- To develop a robust method for high-SNR phase unwrap and line-spectrum target detection.
- To address challenges posed by complex marine ambient noises in DAS.
- To improve the accuracy of underwater acoustic target identification.
Main Methods:
- Proposed a residual module nested long short-term memory (RMN-LSTM) neural network.
- Combined deep spatial feature extraction with long-range temporal dependency.
- Employed a symmetric left-right encoder-decoder network structure.
- Learned direct mapping between distorted wrapped and true phase signals.
Main Results:
- Achieved accurate reconstruction of line-spectrum signals.
- Significantly outperformed conventional algorithms and CNN-based methods.
- Reduced root mean square error (RMSE) to 0.661.
- Obtained a comprehensive SNR improvement of approximately 27 dB.
Conclusions:
- The RMN-LSTM method effectively recovers high-fidelity phase information from noisy DAS data.
- This approach provides a robust technical solution for high-SNR underwater target detection.
- No hardware upgrades are necessary, making the method broadly applicable.
