Bayesian Denoising Algorithm for Low SNR Photon-Counting Lidar Data via Probabilistic Parameter Optimization Based on Signal and Noise Distribution

Qi Liu, Jian Yang, Yue Ma, Wenbo Yu, Qijin Han, Zhibiao Zhou, Song LiRemote 2025 · DOI

TL;DR

This paper introduces a novel Bayesian denoising algorithm for low-SNR photon-counting Lidar data, specifically addressing challenges posed by ICESat-2. By adaptively optimizing parameters based on signal and noise distributions and incorporating MST-based slope estimation, the algorithm achieves robust performance in complex terrains, outperforming existing methods.

What Problem Does This Paper Address?

Photon-counting Lidar, like that used by NASA's ICESat-2, is revolutionary for measuring Earth's surface elevation. However, it faces a significant hurdle: a low signal-to-noise ratio (SNR), especially in bright sunlight or challenging environments. This low SNR makes it difficult to distinguish true surface signals from background noise. Existing denoising methods often rely on fixed parameters, which fail to adapt to the rapidly changing noise and signal characteristics found in rugged mountainous regions. This limitation hinders the full potential of ICESat-2 data for applications like elevation mapping and change detection.

Methods

The core of this work is an adaptive Bayesian denoising framework. It tackles the problem in several key steps:

  1. Simulation Framework: A crucial contribution is a simulation method that generates realistic photon-counting Lidar data with ground truth labels. This is vital because obtaining perfectly labeled real-world data for all scenarios is challenging. The simulation framework, based on ATL03 data, achieves high correlation (>0.9) with measured distributions, providing a reliable benchmark.
  2. MST-Based Feature Extraction: The algorithm uses a Minimum Spanning Tree (MST) approach to extract feature points from the point cloud. This helps in identifying key surface profiles.
  3. Slope Estimation: By fitting lines to these feature points, the algorithm estimates the local surface slope. This slope information is critical for adapting subsequent steps.
  4. Probabilistic Parameter Optimization: Leveraging Bayesian estimation, the algorithm dynamically determines optimal clustering parameters (like search radius and threshold) by analyzing the probability distribution functions (PDFs) of signal and noise photons. This adaptive approach is key to handling varying SNR and terrain.
  5. Elliptic Clustering and Filtering: Finally, an elliptic clustering algorithm, guided by the optimized parameters, performs the denoising. A three-sigma confidence filter is then applied to remove outliers.

Datasets

  • ICESat-2 ATL03 and ATL08 Data: The study utilizes real data from the ICESat-2 mission, specifically focusing on eight tracks from diverse regions (Eastern Utah, Altun Mountain in Tibet, Southern Nevada).
  • Simulated Data: A significant aspect of the work is the creation of a simulation framework that generates synthetic ICESat-2 data. This simulated data, with known ground truth, is used to validate the theoretical model and the denoising algorithm under various controlled SNR and noise conditions. This simulated dataset is a valuable resource for future research.

Key Insights and Ideas

  • Adaptive Parameterization is Crucial: The paper strongly emphasizes that fixed parameters are insufficient for robust Lidar denoising, especially in complex terrains. Dynamic adjustment based on local conditions is paramount.
  • Slope as a Key Indicator: Estimating the local slope using MST is a novel way to inform the denoising process, allowing the algorithm to better handle varying surface orientations.
  • Bayesian Inference for Optimization: Using Bayesian estimation to model signal and noise photon distributions and optimize parameters offers a principled way to handle uncertainty and adapt to changing conditions.
  • Simulation as a Powerful Tool: The development of a high-fidelity simulation framework is a significant contribution, enabling rigorous testing and validation of denoising algorithms where real-world ground truth is scarce.

Key Results

  • High Performance in Low SNR: The proposed algorithm achieves F-scores greater than 0.9 even at extremely low SNR conditions (1 photon/10 MHz noise).
  • Superiority in Challenging Terrain: When compared to benchmark algorithms (OPTICS, Quadtree, DRAGANN) on simulated and real ATL03 datasets, the proposed method demonstrates superior performance in mountainous terrain, showing 10-20% improvements in precision and recall under high noise.
  • Accurate Slope Estimation: The algorithm accurately estimates slopes, with Pearson correlation coefficients generally above 0.9 for single-track data.
  • Robustness to Mislabeled Data: The algorithm shows good robustness to mislabeled signal/noise photons in validation datasets, with computational errors not exceeding 10% for mislabeling ratios below 10%.

Limitations and Future Work

  • Computational Cost: While not explicitly detailed as a limitation, adaptive parameter selection and MST-based processing can be computationally intensive.
  • Generalizability of PDF Model: The paper notes that the underlying methodology of modeling probability density functions for signal and noise photon distributions could be extended to other approaches, such as selecting optimal thresholds in KNN methods or determining optimal pixel sizes in image-based denoising algorithms. This suggests a path for future research to adapt the core ideas to different data types and algorithms.
  • Vegetation Bias: The presence of vegetation can introduce some bias in slope estimation, as noted in the results section. Future work could explore methods to mitigate this specific bias.

Summary updated

Abstract

The Ice, Cloud, and land Elevation Satellite-2 has provided unprecedented global surface elevation measurements through photon-counting Lidar (Light detection and ranging), yet its low signal-to-noise ratio (SNR) poses significant challenges for denoising algorithms. Existing methods, relying on fixed parameters, struggle to adapt to dynamic noise distribution in rugged mountain regions where signal and noise change rapidly. This study proposes an adaptive Bayesian denoising algorithm integrating minimum spanning tree (MST) -based slope estimation and probabilistic parameter optimization. First, a simulation framework based on ATL03 data generates point clouds with ground truth labels under varying SNRs, achieving correlation coefficients > 0.9 between simulated and measured distributions. The algorithm then extracts surface profiles via MST and coarse filtering, fits slopes with >0.9 correlation to reference data, and derives the probability distribution function (PDF) of neighborhood photon counts. Bayesian estimation dynamically selects optimal clustering parameters (search radius and threshold), achieving F-scores > 0.9 even at extremely low SNR (1 photon/10 MHz noise). Validation against three benchmark algorithms (OPTICS, quadtree, DRAGANN) on simulated and ATL03 datasets demonstrates superior performance in mountainous terrain, with precision and recall improvements of 10–20% under high noise conditions. This work provides a robust framework for adaptive parameter selection in low-SNR photon-counting Lidar applications.
BibTeX
@inproceedings{liu2025bayesian,
  author = {Qi Liu and Jian Yang and Yue Ma and Wenbo Yu and Qijin Han and Zhibiao Zhou and Song Li},
  title = {{Bayesian Denoising Algorithm for Low SNR Photon-Counting Lidar Data via Probabilistic Parameter Optimization Based on Signal and Noise Distribution}},
  booktitle = {Remote},
  year = {2025},
  doi = {10.3390/rs17132182},
}

Loading related content...

Figure