Binarization of ESPI fringe patterns based on an M-net convolutional neural network

Area of Science:

• Optics and Photonics
• Computer Vision
• Image Processing

Background:

• Electronic speckle pattern interferometry (ESPI) relies on fringe patterns for phase term estimation.
• Traditional fringe binarization methods struggle with speckle noise and intensity variations inherent in ESPI.
• Accurate binarization is crucial for reliable fringe skeleton extraction and subsequent analysis.

Purpose of the Study:

• To develop an automated and robust method for binarizing challenging ESPI fringe patterns.
• To adapt and improve the M-net convolutional neural network for image segmentation tasks in optical metrology.
• To demonstrate the effectiveness of the proposed method compared to existing techniques.

Main Methods:

• A modified M-net convolutional neural network was employed, treating fringe pattern binarization as a segmentation problem.
• The network was trained using pairs of ESPI fringe patterns and their corresponding ground-truth binary images.
• The modified M-net's performance was evaluated against U-net and RED-net on simulated and experimental ESPI data.

Main Results:

• The proposed M-net method achieved accurate binarization of ESPI fringe patterns, even those with significant noise and intensity inhomogeneity.
• The method successfully produced good results without requiring prior image preprocessing steps.
• Comparative analysis showed the modified M-net outperformed U-net and RED-net in this specific application.

Conclusions:

• The modified M-net convolutional neural network offers an effective solution for robust ESPI fringe pattern binarization.
• This approach enhances the reliability of phase term estimation in ESPI, particularly in noisy conditions.
• The study highlights the potential of deep learning for advanced image processing in optical measurement techniques.