A Feasibility Study of Automated Detection and Classification of Signals in Distributed Acoustic Sensing

Area of Science:

• Marine acoustics
• Signal processing
• Fiber optic sensing

Background:

• Distributed Acoustic Sensing (DAS) offers a novel method for monitoring the marine environment using existing fiber optic infrastructure.
• Near-real-time processing of acoustic data is crucial for effective maritime surveillance and security.

Purpose of the Study:

• To develop and validate an automated framework for detecting and classifying acoustic signals from DAS data.
• To assess the performance of machine learning algorithms (PCA and HDBSCAN) for analyzing spectral characteristics of marine acoustic signals.

Main Methods:

• Utilized data from the SHEFA-2 fiber optic cable for analysis.
• Developed a signal detection and classification framework supporting near-real-time processing.
• Applied Principal Component Analysis (PCA) for data exploration and Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) for classification.

Main Results:

• Achieved a Davies-Bouldin Index of 0.828, Silhouette Score of 0.124, and Calinski-Harabasz Index of 189.8 with the full dataset.
• Demonstrated the capability to distinguish between various acoustic sources including ships, vehicles, earthquakes, and potential cable damage.
• Showcased significant degradation in clustering quality when labeled data was reduced by more than 20%.

Conclusions:

• The developed automated framework shows strong potential for effective maritime acoustic monitoring.
• Maintaining sufficient labeled data is critical for robust classification performance in DAS analysis.
• This technology provides valuable insights for enhancing maritime security and operational awareness.