Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Quantitative Measures for Passive Sonar Texture Analysis

Jarin Ritu, Alexandra Van Dine, Joshua Peeples

TL;DR

The paper addresses the challenge that CNNs struggle to classify passive sonar signals with strong statistical texture. It introduces three synthetic datasets that isolate statistical, structural, and mixed textures, along with two texture quantification metrics: Bidirectional Temporal Entropy (StaTS) and Autocorrelation-based Structural Texture (StrTS). Empirical results show a histogram-based model (HLTDNN) gains substantially over a TDNN baseline on statistically textured and mixed datasets, and mixed-texture pretraining can improve performance on real-world data like DeepShip and VTUAD. The work provides interpretable texture metrics that link data properties to model performance, offering guidance for texture-aware model design in passive sonar and potentially other audio domains.

Abstract

Passive sonar signals contain complex characteristics often arising from environmental noise, vessel machinery, and propagation effects. While convolutional neural networks (CNNs) perform well on passive sonar classification tasks, they can struggle with statistical variations that occur in the data. To investigate this limitation, synthetic underwater acoustic datasets are generated that centered on amplitude and period variations. Two metrics are proposed to quantify and validate these characteristics in the context of statistical and structural texture for passive sonar. These measures are applied to real-world passive sonar datasets to assess texture information in the signals and correlate the performances of the models. Results show that CNNs underperform on statistically textured signals, but incorporating explicit statistical texture modeling yields consistent improvements. These findings highlight the importance of quantifying texture information for passive sonar classification.

Quantitative Measures for Passive Sonar Texture Analysis

TL;DR

The paper addresses the challenge that CNNs struggle to classify passive sonar signals with strong statistical texture. It introduces three synthetic datasets that isolate statistical, structural, and mixed textures, along with two texture quantification metrics: Bidirectional Temporal Entropy (StaTS) and Autocorrelation-based Structural Texture (StrTS). Empirical results show a histogram-based model (HLTDNN) gains substantially over a TDNN baseline on statistically textured and mixed datasets, and mixed-texture pretraining can improve performance on real-world data like DeepShip and VTUAD. The work provides interpretable texture metrics that link data properties to model performance, offering guidance for texture-aware model design in passive sonar and potentially other audio domains.

Abstract

Passive sonar signals contain complex characteristics often arising from environmental noise, vessel machinery, and propagation effects. While convolutional neural networks (CNNs) perform well on passive sonar classification tasks, they can struggle with statistical variations that occur in the data. To investigate this limitation, synthetic underwater acoustic datasets are generated that centered on amplitude and period variations. Two metrics are proposed to quantify and validate these characteristics in the context of statistical and structural texture for passive sonar. These measures are applied to real-world passive sonar datasets to assess texture information in the signals and correlate the performances of the models. Results show that CNNs underperform on statistically textured signals, but incorporating explicit statistical texture modeling yields consistent improvements. These findings highlight the importance of quantifying texture information for passive sonar classification.

Paper Structure

This paper contains 14 sections, 9 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Statistical Texture Dataset
  4. Structural Texture Dataset
  5. Mixed Statistical and Structural Textures Dataset
  6. Statistical Texture Quantification via Bidirectional Temporal Entropy
  7. Structural Texture Quantification via Autocorrelation
  8. Experimental Procedure
  9. Dataset Description
  10. Experimental Design
  11. Results and Discussion
  12. Synthetic and Real-World Datasets Evaluation
  13. Pretraining Models Using Synthetic Datasets
  14. Conclusion

Figures (2)

  • Figure 1: Example spectrograms showing one sample per class (Vessel 1, Vessel 2, Vessel 3, Vessel 4) for each synthetic dataset type: Structural Texture, Statistical Texture, and Mixed Statistical and Structural Textures.
  • Figure 2: Self-similarity matrices for three representative signals. (a) White noise (StrTS = 0.05): lacks clear structure. (b) Cargo ship signal (StrTS = 2.73): shows partial diagonal patterns. (c) Pure sine wave (StrTS = 5.00): exhibits strong and evenly spaced diagonals.