Segmentation Method for Ship-Radiated Noise Using the Generalized Likelihood Ratio Test on an Ordinal Pattern...

Area of Science:

• Signal Processing
• Acoustics
• Machine Learning

Background:

• Ship-radiated noise (SRN) segmentation is crucial for identifying ship statuses.
• Existing methods struggle with SRN diversity due to a lack of prior knowledge.
• Acoustic feature extraction and probability distribution estimation pose challenges for SRN analysis.

Purpose of the Study:

• To develop a robust audio segmentation method for ship-radiated noise (SRN).
• To improve the accuracy of single change-point detection (SCPD) and multiple change-points detection (MCPD) in SRN.
• To provide a more effective segmentation criterion for ship classification.

Main Methods:

• Proposed a segmentation criterion based on a generalized likelihood ratio (GLR) test applied to ordinal pattern distribution (OPD).
• Introduced the GLR-OPD method for single change-point detection (SCPD) and multiple change-points detection (MCPD).
• Efficiently estimated OPD using the sequential structure of ordinal patterns across analysis windows, avoiding feature extraction and distribution estimation.

Main Results:

• The proposed method accurately estimates the number and location of change-points in synthetic signals.
• Segmentation results on real-world SRN demonstrate more distinguishable segments compared to Bayesian Information Criterion (BIC) methods.
• The method proves effective for SRN segmentation, enhancing subsequent classification.

Conclusions:

• The GLR-OPD method offers a significant advancement in ship-radiated noise segmentation.
• This approach overcomes limitations of existing methods by not requiring prior knowledge or complex feature extraction.
• The enhanced segmentation capability directly contributes to more accurate ship status and category identification.