A New Feature Extraction Method Based on Improved Variational Mode Decomposition, Normalized Maximal Information...

Area of Science:

• Underwater acoustics
• Signal processing
• Machine learning

Background:

• Marine environmental noise and unstable underwater acoustic channels corrupt ship-radiated noise (SRN) signals.
• Signal denoising is crucial for accurate underwater acoustic target recognition.

Purpose of the Study:

• To develop a novel hybrid feature extraction scheme for denoising and recognizing SRN signals.
• To improve the accuracy of underwater acoustic target identification.

Main Methods:

• Improved variational mode decomposition (IVMD) to decompose SRN signals into intrinsic mode functions (IMFs).
• Filtering of noise IMFs, followed by permutation entropy (PE) extraction.
• Normalized maximal information coefficient (norMIC) to weigh PE values of retained IMFs.
• Particle swarm optimization-based support vector machine (PSO-SVM) classifier for SRN sample identification.

Main Results:

• The proposed hybrid method achieved a classification accuracy of 99.1667%.
• Demonstrated significantly higher accuracy compared to existing methods.
• The feature extraction scheme is effective for practical applications.

Conclusions:

• The integrated IVMD, norMIC, and PE approach provides a robust solution for SRN signal feature extraction.
• The method enhances the performance of underwater acoustic target recognition systems.
• The proposed technique is suitable for real-world applications requiring accurate SRN signal identification.