A Robust Denoised Algorithm Based on Hessian–Sparse Deconvolution for Passive Underwater Acoustic Detection

Fan Yin, Chao Li, Haibin Wang, Shihong Zhou, Leixin Nie, Yonglin Zhang, Hao YinJournal of Marine Science and Engineering 2023 · DOI

TL;DR

This paper introduces a novel Hessian-sparse deconvolution algorithm for underwater acoustic detection. By adapting image processing techniques, it significantly improves azimuthal resolution and reduces noise in bearing time records (BTRs), especially in challenging low-SNR environments, outperforming traditional methods.

What Problem Does This Paper Address?

Underwater acoustic array signal processing is crucial for detecting and tracking targets. However, a major limitation is the "Rayleigh limit," which restricts the achievable azimuthal resolution, making it difficult to distinguish closely spaced targets or detect faint ones in noisy ocean environments. Traditional beamforming methods like CBF and MVDR, while robust, suffer from this resolution constraint and are further hampered by complex ocean noise (biological, ship, hydrodynamic), leading to low signal-to-noise ratios (SNRs). While deconvolution techniques have been explored, existing methods often struggle with convergence in low SNRs or don't fully exploit the potential for improved resolution. This paper aims to overcome these limitations by developing a more effective post-processing algorithm that enhances both resolution and noise suppression.

Methods

The core of this work is a novel two-dimensional Hessian-sparse deconvolution algorithm. It builds upon existing deconvolution techniques but introduces several key innovations:

  • Two-Dimensional Deconvolution: Unlike previous one-dimensional approaches that primarily focus on spatial resolution, this method extends deconvolution to two dimensions, processing both the spatial (bearing) and temporal dimensions of the BTR. This allows for additional processing gain and helps address issues like motion blur.
  • Hessian Matrix Regularization: The algorithm incorporates a Hessian matrix regularization term. This term, derived from image processing, helps preserve sharp edges while mitigating the "staircase effect" often seen in Total Variation (TV) denoising methods, leading to smoother and more natural-looking results.
  • Sparsity Constraint: Recognizing that target trajectories in BTRs often exhibit sparsity, a sparsity-norm constraint is added. This helps to further refine the deconvolution process and suppress unwanted noise and scattered signals.
  • Split Bregman Algorithm: To efficiently solve the complex optimization problem arising from the Hessian-sparse deconvolution, the authors employ the split Bregman algorithm. This iterative method is known for its speed and effectiveness in image restoration tasks.
  • Integration with Image Processing: The approach leverages concepts and techniques from image processing, treating BTRs as images and applying methods like partial differential equation denoising and sparse reconstruction.

The overall workflow involves calculating the Hessian matrix, performing 2D deconvolution, generating a Point Spread Function (PSF), and then applying iterative deconvolution to refine the BTR.

Datasets

The paper utilizes two types of data for evaluation:

  • Simulated Data: The authors generated simulated underwater acoustic data to test the algorithm's performance under controlled conditions, specifically focusing on low-SNR environments and multi-target scenarios with cross-interference.
  • Real Sea Trial Data: Actual data collected from experiments in the South China Sea was used to validate the algorithm's practical applicability in a complex marine environment. This data was acquired using a towed horizontal array.

No new benchmark datasets were introduced.

Key Insights and Ideas

  • Bridging Image Processing and Acoustics: The paper successfully demonstrates that techniques from image processing, particularly those dealing with noise reduction and resolution enhancement (like Hessian regularization and sparsity constraints), can be highly effective when adapted for underwater acoustic signal processing.
  • Beyond 1D Deconvolution: Extending deconvolution to two dimensions (spatial and temporal) offers significant advantages, not just for spatial resolution but also for handling temporal aspects like motion blur, which is crucial for dynamic underwater targets.
  • Hessian for Smoother Denoising: The use of Hessian matrix regularization provides a better balance between noise suppression and edge preservation compared to traditional TV methods, avoiding the "staircase effect" and producing more visually appealing results.
  • Sparsity as a Prior: Incorporating sparsity constraints leverages the inherent characteristics of target trajectories in BTRs, leading to more accurate reconstruction and better suppression of spurious signals.
  • Robustness in Low SNR: The proposed method shows remarkable performance even in very low SNR conditions (-20 dB and below), where traditional methods struggle significantly.

Key Results

  • Main Lobe Width Reduction: The proposed Hessian-sparse deconvolution method achieved a main lobe width of approximately 1.3 degrees at 0 dB SNR, a significant improvement over CBF (6.0 degrees) and 1D deconvolution (3.9 degrees). This reduction was maintained across various SNR levels, with the proposed method consistently outperforming others.
  • Noise Suppression: In low-SNR environments (e.g., -20 dB), the algorithm demonstrated substantial background noise reduction, approximately 10 dB better than the original BTR and significantly more than 1D deconvolution.
  • Multi-Target Resolution: In multi-target scenarios, the algorithm effectively distinguished between closely spaced targets, even when their initial energy levels were comparable. It showed more pronounced main lobe contraction for fast-moving targets, addressing temporal aspects.
  • Sea Trial Validation: Real sea trial data confirmed the algorithm's effectiveness in suppressing complex marine background noise and narrowing main lobes, leading to clearer target detection, especially for weaker targets that were missed by traditional methods.

Limitations and Future Work

  • Suppression of Weak Targets: The paper acknowledges that the introduction of sparsity constraints, while beneficial for overall noise reduction, can sometimes lead to the suppression of very weak targets, especially those scattered in specific angular regions.
  • Parameter Sensitivity: The instability observed in some denoising and main lobe contraction processes is attributed to the settings of prior knowledge parameters and the choice of optimization iteration methods.
  • Information Loss/Over-Denoising: In some cases, the method might lead to a loss of local information or over-denoising.

Future work suggested by the authors includes:

  • Exploring Generic Prior Knowledge: Investigating strategies to construct more accurate and generic prior knowledge to improve the robustness of the algorithm.
  • Avoiding Excessive Denoising: Developing methods to prevent over-denoising and better preserve subtle target information.
  • Optimizing Parameter Settings: Further research into optimizing the parameters of the optimization algorithms and regularization terms.

Summary updated

Abstract

Digital beamforming techniques find wide applications in the field of underwater acoustic array signal processing. However, their azimuthal resolution has long been constrained by the Rayleigh limit, consequently limiting their detection performance. In this paper, we propose a novel two-dimensional Hessian–sparse deconvolution algorithm based on image processing techniques. This method assumes a priori that the underwater acoustic bearing time record (BTR) images exhibit sparsity, and then it first constructs partial differential equations in the beamforming domain with sparsity-norm constraints for optimal noise reduction. Subsequently, a two-dimensional deconvolution operation is applied to narrow the main lobe, aiming to achieve additional temporal gains in two-dimensional processing. The simulation and real sea trial data processing results show that the main lobe width of the proposed method is about 1.3 degrees at 0 dB. It effectively reduces the main lobe width and enhances the detection resolution of BTRs in the post-processing part, especially in low-signal-to-noise-ratio (SNR) environments. Therefore, the proposed method provides nice opportunities to further improve the target-detecting ability of hydrophone arrays.
BibTeX
@article{yin2023robust,
  author = {Fan Yin and Chao Li and Haibin Wang and Shihong Zhou and Leixin Nie and Yonglin Zhang and Hao Yin},
  title = {{A Robust Denoised Algorithm Based on Hessian–Sparse Deconvolution for Passive Underwater Acoustic Detection}},
  journal = {Journal of Marine Science and Engineering},
  year = {2023},
  doi = {10.3390/jmse11102028},
}
Figure