Information Complexity of Time-Frequency Distributions of Signals in Detection and Classification Problems

Area of Science:

• Signal Processing
• Machine Learning
• Acoustics

Background:

• Acoustic signal detection and classification are crucial in various fields.
• Traditional methods may lack robustness or require extensive feature engineering.
• Information-theoretic approaches offer a promising alternative for signal analysis.

Purpose of the Study:

• To develop and evaluate new information-based features for acoustic signal detection and classification.
• To assess the efficacy of these features using machine learning on hydroacoustic data.

Main Methods:

• Proposed novel information features derived from time-frequency distributions, including spectrograms and reassigned spectrograms.
• Employed machine learning algorithms for multiclass classification.
• Validated methods using synthetic signal modeling and real hydroacoustic recordings.

Main Results:

• The proposed information features demonstrated strong performance in classifying acoustic signals.
• Achieved a high classification accuracy with an F1 score of 0.95 on real hydroacoustic data.
• The reassigned spectrogram proved to be a valuable enhancement for feature extraction.

Conclusions:

• The developed information-based features are effective for acoustic signal detection and classification.
• The proposed approach offers advantages over existing methods, particularly for hydroacoustic applications.
• Further research can explore these features in other acoustic domains.