A laser radar point cloud denoising method based on density-aware deformable attention and adversarial domain adaptive transfer learning

By combining density-aware deformable attention mechanism and adversarial domain adaptive transfer learning, the problems of poor adaptability of traditional methods and reliance on a large amount of labeled data in LiDAR point cloud data processing are solved. High-precision denoising is achieved in scenarios where labeled data is scarce, thus improving the processing effect of point cloud data.

CN122243800APending Publication Date: 2026-06-19GUILIN UNIVERSITY OF TECHNOLOGY
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202610551973.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-04-24
Publication Date
2026-06-19

AI Technical Summary

Technical Problem

In the processing of lidar point cloud data, traditional methods have poor adaptability and limited generalization ability, while deep learning methods rely heavily on a large amount of real labeled data, making it difficult to achieve high-precision noise reduction in scenarios such as spaceborne environments where labeled data is scarce.

Method used

By employing a density-aware deformable attention mechanism and an adversarial domain adaptive transfer learning approach, a feature extraction mechanism capable of sensing the local density distribution of point clouds and adaptively adjusting the region of interest is designed. Combined with an adversarial transfer learning strategy, the difference between simulated and real data distributions is reduced, achieving high-precision and strong-generalization denoising.

Benefits of technology

With only a small amount of real labeled data, high-precision denoising of LiDAR point clouds was achieved, improving the model's generalization ability on real data and enhancing the processing accuracy and reliability of point cloud data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122243800A_ABST
    Figure CN122243800A_ABST
Patent Text Reader

Abstract

This invention relates to the field of lidar point cloud data processing technology, specifically disclosing a lidar point cloud denoising method based on density-aware deformable attention and adversarial domain adaptive transfer learning. This method aims to address the problems of poor adaptability and limited generalization ability of traditional denoising methods, as well as the heavy reliance of deep learning methods on large amounts of real labeled data. Its core lies in extracting noise-robust local features through a density-aware deformable attention mechanism and using an adversarial domain adaptive transfer learning strategy to reduce the domain difference between simulated and real data. This invention employs a two-stage training strategy: first, the model is pre-trained on simulated data, and then fine-tuned using a very small amount of real labeled data combined with adversarial learning. This method is particularly suitable for scenarios with scarce labeled data (such as spaceborne platforms), and can achieve high-precision, strong-generalization denoising of lidar point clouds under limited training data conditions, comprehensively outperforming traditional methods in key performance indicators.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • ICESat-2 mountainous area surface laser point cloud extraction method

    CN116109751A

  • Denoising method for photon cloud data of different beams under daytime condition of multistage filtering algorithm

    CN116165635A

  • Photon counting laser radar denoising method based on improved DENCLUE

    CN119887577A

  • Laser radar photon point cloud denoising method for lake water level monitoring

    CN120198321A

  • Point cloud static denoising method based on geometric constraint and deep learning

    CN121032850A