A three-Stage training strategy phase unwrapping method for high speckle noises DOI Creative Commons

Kejia Li,

Zixin Zhao, Hong Zhao

et al.

Optics Express, Journal Year: 2024, Volume and Issue: 32(27), P. 48895 - 48895

Published: Dec. 6, 2024

Deep learning has been widely used in phase unwrapping. However, owing to the noise of wrapped phase, errors wrap count prediction and calculation can occur, making it challenging achieve high measurement accuracy under high-noise conditions. To address this issue, a three-stage multi-task unwrapping method was proposed. The retrieval divided into three training stages: denoising, prediction, unwrapped error compensation. In first stage, preprocessing module trained reduce interference, thereby improving calculation. second stage involved module. A residual compensation added correct from denoising results generated stage. Finally, third calculated Additionally, convolution-based multi-scale spatial attention proposed, which effectively reduces interference spatially inconsistent be applied convolutional neural network. principles based on strategy were introduced. Subsequently, framework strategies for each presented. tested using simulated data with varying levels. It compared TIE, iterative least squares method, UNet, phaseNet2.0, DeepLabV3 + correction operation, demonstrating robustness proposed method.

Language: Английский

DenSFA-PU: Learning to unwrap phase in severe noisy conditions DOI Creative Commons
M.M. Awais, Taeil Yoon,

Cholsong Hwang

et al.

Optics & Laser Technology, Journal Year: 2025, Volume and Issue: 187, P. 112757 - 112757

Published: March 12, 2025

Language: Английский

Citations

1
ATCM-Net: A deep learning method for phase unwrapping based on perception optimization and learning enhancement DOI
Min Xu,

Jia Cong,

Yuxin Shen

et al.

Optics & Laser Technology, Journal Year: 2025, Volume and Issue: 190, P. 113185 - 113185

Published: May 19, 2025

Language: Английский

Citations

0
Design and Analysis of Orthogonal Polarization Point Diffraction Pinhole Plate DOI Creative Commons

Ziyu Han,

Wenlu Feng,

Zhilin Zhang

et al.

Photonics, Journal Year: 2024, Volume and Issue: 11(7), P. 602 - 602

Published: June 26, 2024

The pinhole plate is a key component of the point diffraction interferometer (PDI). reasonable improvement and simulation this device would enhance application interferometry technology during measurement wavefronts. traditional method easily disturbed by environmental noise, making it difficult to obtain high-precision dynamic measurements. This paper introduces four-step phase-shift PDI that can be employed in common optical path. By using principle finite-difference time-domain (FDTD), model orthogonal polarization (OP-PDPP) structure established. results show when Cr used as film material plate, parameters include thickness 150 nm, diameter 2 μm, wire grid period width 100 nm; addition, comprehensive extinction ratio greatest wavefront error smallest. Finally, constructed experimental system test flat sample with 25.4 mm aperture, are compared those ZYGO interferometer. difference peak-to-valley (PV) value between OP-PDI 0.0028λ, an RMS 0.0011λ; verifies feasibility scheme proposed paper. OP-PDPP effective tool for measurement.

Language: Английский

Citations

1
End-to-end color fringe depth estimation based on a three-branch U-net network DOI

Xinjun Zhu,

Tianyang Lan,

Yixin Zhao

et al.

Applied Optics, Journal Year: 2024, Volume and Issue: 63(28), P. 7465 - 7465

Published: Sept. 9, 2024

In fringe projection profilometry (FPP), end-to-end depth estimation from patterns for FPP attracts more and attention patterns. However, color images provide additional information the RGB channel FPP, which has been paid little in estimation. To this end, paper we present first time, to best of our knowledge, an network using composite fringes with better performance. order take advantage pattern, a multi-branch structure is designed paper, learns multi-channel details object under test by three encoders each introduces module capture complex features modalities input data. Experiments simulated real datasets show that proposed method pattern effective estimation, it outperforms other deep learning methods such as UNet, R2Unet, PCTNet, DNCNN.

Language: Английский

Citations

1
PUDCN: two-dimensional phase unwrapping with a deformable convolutional network DOI Creative Commons

Youxing Li,

Lingzhi Meng, Kai Zhang

et al.

Optics Express, Journal Year: 2024, Volume and Issue: 32(16), P. 27206 - 27206

Published: July 3, 2024

Two-dimensional phase unwrapping is a fundamental yet vital task in optical imaging and measurement. In this paper, what we believe to be novel deep learning framework PUDCN proposed for 2D unwrapping. We introduce the deformable convolution technique design two convolution-related plugins dynamic feature extraction. addition, adopts coarse-to-fine strategy that unwraps first stage then refines unwrapped second obtain an accurate result. The experiments show our performs better than existing state-of-the-art. Furthermore, apply unwrap of fibers interferometry, demonstrating its generalization ability.

Language: Английский

Citations

1
High-Accuracy Phase Unwrapping Based on Binarized Wrap Count DOI Creative Commons
Huazhen Liu, Rongjun Shao, Yuan Qu

et al.

Optics Express, Journal Year: 2024, Volume and Issue: 32(25), P. 44605 - 44605

Published: Nov. 12, 2024

Spatial phase unwrapping is essential for converting wrapped fringes into a continuous unwrapped map, which critical various high-precision measurement technologies. The accuracy of directly affects precision. Recently, deep learning-based has emerged as promising alternative to traditional methods, primarily due its strong resilience against noise. However, existing approaches often struggle produce consistent results, limiting their practical applicability. This study introduces binarized wrap count (BWCPU), we belive novel method that utilizes neural networks analyze gradient structures through counts. approach reduces prediction complexity while ensuring accurate segmentation. In structured light surface measurements, BWCPU significantly decreases misinterpretations in noisy conditions, achieving remarkable 76.9% improvement over leading wrap-count estimation methods. Furthermore, by employing stitching algorithm known unidirectional optimal seam stitching, extends capabilities handle 1024 × patterns, showcasing potential measurements environments.

Language: Английский

Citations

0
A three-Stage training strategy phase unwrapping method for high speckle noises DOI Creative Commons

Kejia Li,

Zixin Zhao, Hong Zhao

et al.

Optics Express, Journal Year: 2024, Volume and Issue: 32(27), P. 48895 - 48895

Published: Dec. 6, 2024

Deep learning has been widely used in phase unwrapping. However, owing to the noise of wrapped phase, errors wrap count prediction and calculation can occur, making it challenging achieve high measurement accuracy under high-noise conditions. To address this issue, a three-stage multi-task unwrapping method was proposed. The retrieval divided into three training stages: denoising, prediction, unwrapped error compensation. In first stage, preprocessing module trained reduce interference, thereby improving calculation. second stage involved module. A residual compensation added correct from denoising results generated stage. Finally, third calculated Additionally, convolution-based multi-scale spatial attention proposed, which effectively reduces interference spatially inconsistent be applied convolutional neural network. principles based on strategy were introduced. Subsequently, framework strategies for each presented. tested using simulated data with varying levels. It compared TIE, iterative least squares method, UNet, phaseNet2.0, DeepLabV3 + correction operation, demonstrating robustness proposed method.

Language: Английский

Citations

0