Robust ESPI fringe pattern denoising method using the Restormer Partial Differential Network (Res-PDNet).

Area of Science:

• * Applied Physics
• * Optical Engineering
• * Computer Vision

Background:

• * Electronic Speckle Pattern Interferometry (ESPI) is a crucial non-destructive testing (NDT) method.
• * Denoising interference fringe patterns is a critical and challenging aspect of ESPI.
• * Existing denoising methods often struggle to balance noise reduction with fringe structure preservation.

Purpose of the Study:

• * To develop an advanced deep learning model for effective denoising of ESPI fringe patterns.
• * To enhance the accuracy and reliability of ESPI by improving fringe pattern quality.
• * To preserve the integrity of fringe structures and shapes during the denoising process.

Main Methods:

• * Introduction of the Restormer Partial Differential Network (Res-PDNet), integrating partial differential equation (PDE) denoising with deep learning.
• * Incorporation of Restormer's Multi-Dconv Head Transposed Attention and Gated-Dconv Feed-Forward Network modules into the PDNet architecture.
• * Integration of orientation constraints within the loss function to preserve fringe pattern geometry.

Main Results:

• * The Res-PDNet effectively filters noise from ESPI fringe patterns.
• * The proposed method demonstrates superior performance in preserving fringe structure and shape compared to traditional methods.
• * Accurate recognition of fringe patterns and effective noise removal were achieved on electron scattering interference patterns.

Conclusions:

• * Res-PDNet offers a robust solution for ESPI fringe pattern denoising.
• * The enhanced network architecture and loss function contribute to improved fringe preservation.
• * This technique holds significant potential for advancing non-destructive testing applications using ESPI.