A Bayesian optimization framework for two-dimensional sparse array design driven by physics-based simulation.

Area of Science:

• Underwater acoustics
• Array signal processing
• Acoustic imaging

Background:

• Traditional underwater array design prioritizes local metrics, neglecting overall imaging performance in real-world acoustic conditions.
• Existing methods lack optimization for practical acoustic environments, limiting imaging capabilities.

Purpose of the Study:

• To propose a collaborative array design framework integrating physical acoustic simulation and Bayesian optimization for enhanced underwater imaging.
• To develop an automatic search for optimal array configurations using a Gaussian process surrogate model and a dual-stage delay optimization strategy.

Main Methods:

• Implemented a collaborative design framework combining physical acoustic simulation with Bayesian optimization.
• Utilized a Gaussian process as a surrogate model within a dual-stage delay optimization (DSDO) strategy.
• Employed a composite loss function for automatic optimization of array configurations.

Main Results:

• The DSDO method reduced maximum delay mean square error by approximately 66.7% compared to the Fresnel approximation within a 90° × 90° field of view.
• Optimized arrays showed significant improvements in image segmentation metrics: PSNR (+6.49%), SSIM (+1.62%), IoU (+6.21%), and accuracy (+2.17%) versus the Fermat spiral array.
• Enhanced clarity and structural fidelity of targets within acoustic images were demonstrated.

Conclusions:

• The proposed DSDO method effectively optimizes underwater acoustic array design for improved imaging performance.
• The framework provides a foundation for enhanced target detection and recognition in practical acoustic environments.
• This approach addresses limitations in traditional array design by focusing on holistic imaging quality.