A New Feature Extraction Method for Ship-Radiated Noise Based on Improved CEEMDAN, Normalized Mutual Information and...

Area of Science:

• Signal Processing
• Acoustics
• Machine Learning

Background:

• Passive sonar performance relies on effective feature extraction from ship-radiated noise.
• Existing entropy-based methods for ship classification are unreliable in noisy conditions due to lack of noise reduction and single-scale limitations.

Purpose of the Study:

• To develop a robust and reliable feature extraction method for ship-radiated noise.
• To improve ship classification performance under varying noise conditions.

Main Methods:

• Utilized improved complementary ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) to decompose ship-radiated noise into intrinsic mode functions (IMFs).
• Implemented a noise reduction process by identifying and removing noise-dominant IMFs.
• Calculated normalized mutual information (norMI) and multiscale improved permutation entropy (MIPE) for signal-dominant IMFs, with norMI weighting MIPE results to create a multi-scale entropy feature.

Main Results:

• The proposed method achieved a recognition rate of 90.67% in noise-free conditions and 83% under 5 dB noise conditions.
• Demonstrated significantly higher accuracy compared to existing entropy feature extraction algorithms.

Conclusions:

• The novel feature extraction method is more reliable and suitable for practical application in ship-radiated noise analysis.
• The integration of ICEEMDAN, norMI, and MIPE effectively addresses the limitations of previous entropy-based approaches.