Multi-Stage Feature Extraction and Classification for Ship-Radiated Noise.

Area of Science:

• Signal Processing
• Acoustics
• Machine Learning

Background:

• Ship-radiated noise (SRN) detection via passive sonar is often distorted by complex hydroacoustic channels.
• Existing SRN feature extraction methods have significant limitations, hindering accurate signal analysis.
• Effective feature extraction is crucial for enhancing the performance of passive sonar systems.

Purpose of the Study:

• To propose a novel, multi-stage feature extraction approach for improving SRN detection and classification.
• To address the shortcomings of current SRN feature extraction techniques.
• To enhance the accuracy and reliability of passive sonar signal analysis.

Main Methods:

• Utilizing enhanced variational mode decomposition (EVMD) to decompose SRN signals into intrinsic mode functions (IMFs).
• Calculating weighted permutation entropy (WPE) for each IMF to extract features.
• Applying local tangent space alignment (LTSA) for dimensionality reduction of extracted features.
• Employing particle swarm optimization-based support vector machine (PSO-SVM) for multi-class classification of SRN samples.

Main Results:

• The proposed multi-stage method effectively extracts features from complex SRN signals.
• Dimensionality reduction using LTSA preserves crucial information for classification.
• The PSO-SVM classifier achieved a high recognition rate of up to 96.6667% for different SRN types.
• Simulation and experimental results confirm the superiority over conventional SRN feature extraction methods.

Conclusions:

• The proposed EVMD-WPE-LTSA-PSO-SVM approach offers a significant advancement in SRN feature extraction.
• This method enhances the performance of passive sonar systems in challenging hydroacoustic environments.
• The high recognition rate demonstrates the practical applicability and effectiveness of the proposed technique.