A two-stage few-shot learning-based paired particle image denoising method

The pairwise particle image denoising method based on two-stage few-sample learning solves the problem of inconsistent particle features between frames in complex noise environments, achieving high-quality particle image denoising and improved velocity field measurement accuracy. It is suitable for complex flow measurements in experimental fluid mechanics and wind engineering.

CN122115264APending Publication Date: 2026-05-29HARBIN INST OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
HARBIN INST OF TECH
Filing Date
2026-03-06
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing particle image velocimetry methods struggle to achieve high-quality pairwise particle image denoising in complex noise environments, leading to inconsistencies in particle features between frames and affecting the accuracy and robustness of velocity field measurements.

Method used

A pairwise particle image denoising method based on two-stage few-shot learning is adopted. By constructing a large-scale synthetic particle image dataset and a noisy-clean pairwise particle image dataset, and combining a time-adaptive feature alignment module, a noise-aware residual feature modulation module, and a shared feature refinement and reconstruction module, a deep neural network is trained to achieve frame-consistent denoising.

Benefits of technology

It significantly improves image quality and velocity field measurement accuracy with a small number of samples, effectively suppresses complex noise interference, maintains particle shape and edge sharpness, and improves the robustness of PIV measurement.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122115264A_ABST
    Figure CN122115264A_ABST
Patent Text Reader

Abstract

The application discloses a two-stage few-shot learning-based paired particle image denoising method, and belongs to the technical field of experimental fluid mechanics and image processing. First, a large-scale synthetic particle image dataset and a noisy-clean paired particle image dataset containing various noise types are constructed. Then, a deep neural network containing a time-adaptive feature alignment module, a noise-aware residual feature modulation module and a shared feature refinement reconstruction module is constructed. A two-stage training strategy is adopted. In the first stage, the feature alignment and reconstruction modules are pre-trained on the large-scale synthetic dataset. In the second stage, the noise-aware modulation module is fine-tuned on a small number of noisy-clean samples. Finally, the particle image pair to be denoised is read, and the trained model is used to process the particle image pair to obtain a high-quality denoised image pair. The application can realize high-quality denoising with a small number of samples, ensures inter-frame consistency through time-adaptive alignment, and significantly improves the PIV velocity field measurement accuracy and robustness.
Need to check novelty before this filing date? Find Prior Art