Underwater single-channel acoustic signal multitarget recognition using convolutional neural networks

Area of Science:

• Underwater acoustics
• Signal processing
• Machine learning

Background:

• Ship radiated noise is crucial for underwater target recognition.
• Existing deep learning methods primarily focus on single-target recognition.
• Multitarget underwater acoustic signal recognition remains a challenge.

Purpose of the Study:

• To propose a deep learning-based method for single-channel multitarget underwater acoustic signal recognition.
• To address the recognition of both unique and duplicate categories within multiple targets.
• To evaluate the effectiveness of different input features for multitarget recognition.

Main Methods:

• Utilized real-valued and complex-valued ResNet and DenseNet convolutional neural networks.
• Generated synthetic multitarget signals by superimposing individual target signals.
• Compared performance using original audio, complex-valued STFT, magnitude STFT, log-mel spectrum, and MFCCs as input features.

Main Results:

• The proposed deep learning method successfully recognized synthetic multitarget ship signals.
• Effective recognition was achieved using magnitude STFT spectrum, complex-valued STFT spectrum, and log-mel spectrum as inputs.
• The method demonstrated capability in solving both multilabel binary and multilabel multiple value classification tasks.

Conclusions:

• The developed deep learning approach offers a viable solution for multitarget underwater acoustic signal recognition.
• Specific spectral features significantly enhance the performance of multitarget recognition systems.
• This research advances the capability of identifying multiple underwater acoustic targets from single-channel data.