A Deep Learning-Based Method for Reconstructing the 3D Position of Color-Copied Particles

CN122134918APending Publication Date: 2026-06-02NANJING UNIV OF AERONAUTICS & ASTRONAUTICS

Patent Information

Authority / Receiving Office: CN · China
Patent Type: Applications(China)
Current Assignee / Owner: NANJING UNIV OF AERONAUTICS & ASTRONAUTICS
Filing Date: 2026-01-15
Publication Date: 2026-06-02

AI Technical Summary

Technical Problem

Existing 3D particle reconstruction methods are highly dependent on hardware, costly, and the reconstruction results are unstable when the number of particles increases or when there is occlusion, making it difficult to maintain high accuracy under different imaging conditions.

Method used

A deep learning-based color coding method is adopted. By acquiring two-dimensional color particle images and point spread functions, a deep learning network is constructed. The U-Net generator and decoder are used to reconstruct the three-dimensional coordinate point cloud of particles. The matching filter response and color coding features are combined to improve the reconstruction accuracy and stability.

Benefits of technology

While reducing hardware costs, it improves the accuracy and stability of 3D particle reconstruction, especially maintaining good reconstruction results under complex imaging conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

Figure CN122134918A_ABST

Patent Text Reader

Abstract

This invention discloses a deep learning-based method for reconstructing the 3D position of color-coded particles, comprising the following steps: Step 1, acquiring a two-dimensional color particle image and its corresponding point spread function (PSF), as well as particle position coordinates; Step 2, constructing a training set of the two-dimensional color particle image, PSF matched filter response, and particle position coordinate point cloud: obtaining the response of the two-dimensional color particle image to PSF at different depths based on the matched filter, and constructing a pair of data in the training set with the two-dimensional color particle image and the corresponding particle 3D position point cloud; Step 3, constructing and training a deep learning network: constructing a U-Net-based deep learning network, concatenating the matched filter response to the network input, and performing mapping training; Step 4, using the trained deep learning network to reconstruct the particle 3D coordinate point cloud from the captured two-dimensional color particle image.

Need to check novelty before this filing date? Find Prior Art