Speckle denoising based on deep learning via a conditional generative adversarial network in digital holographic...

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Metrology

Background:

• Speckle noise in digital holographic interferometry (DHI) can degrade phase measurement accuracy.
• Existing denoising methods may compromise experimental data integrity.

Purpose of the Study:

• To develop a deep learning-based speckle denoising algorithm for DHI.
• To improve the accuracy and efficiency of phase measurements in DHI.

Main Methods:

• A conditional generative adversarial network (cGAN) was employed, utilizing U-Net and DenseNet architectures for discriminator and generator subnetworks.
• Speckle simulation datasets were used for training to enhance noise feature extraction.
• A loss function incorporating peak signal-to-noise ratio (PSNR) was designed for improved performance.

Main Results:

• The proposed deep learning algorithm effectively extracts noise features from simulated speckle data.
• The cGAN-based method demonstrated superior speckle denoising compared to other algorithms.
• The algorithm improved the accuracy of experimental strain data processing from DHI.

Conclusions:

• Deep learning, specifically cGANs, offers a powerful approach for speckle denoising in DHI.
• The developed algorithm enhances phase measurement accuracy and preserves experimental data integrity.
• This method shows significant potential for applications involving DHI strain analysis.