Enhanced Doppler tomography LiDAR imaging using the MACE framework and deep denoisers.

Area of Science:

• Optical remote sensing
• Signal processing
• Aerospace engineering

Background:

• Doppler tomographic LiDAR (DTL) enables long-range imaging of rotating targets for applications like drone surveillance and space debris tracking.
• DTL performance is significantly degraded by shot noise and non-stationary speckle artifacts, limiting its operational range and imaging quality.

Purpose of the Study:

• To develop a novel DTL reconstruction framework that overcomes inherent noise limitations for improved imaging of rapidly rotating targets.
• To enhance the practical viability of DTL in realistic, noisy environments for critical aerospace applications.

Main Methods:

• A multi-agent consensus equilibrium (MACE) framework integrating a refined physical imaging model with data-driven denoising.
• Development of a Bayesian-optimized physical model to characterize noise propagation dynamics.
• Implementation of two specialized deep denoisers targeting shot noise and speckle noise, harmonized via MACE for adaptive noise suppression.

Main Results:

• The proposed MACE-based DTL framework significantly outperforms state-of-the-art techniques, particularly under low signal-to-noise ratio (SNR) conditions.
• Successful imaging of a target rotating at 2,400 RPM from 40 meters with an SNR of -10 dB, where other methods failed.
• Demonstrated robustness and adaptability across diverse noise environments through simulations and real-world experiments.

Conclusions:

• The developed MACE framework offers a flexible and robust solution for high-resolution imaging of fast-rotating targets in realistic noise conditions.
• This advancement significantly improves the practical applicability of DTL for crucial aerospace surveillance and characterization tasks.