A convolutional neural network-based framework for quality control through speckle displacement analysis.

Area of Science:

• Optical Metrology
• Image Processing
• Machine Learning

Background:

• Speckle metrology relies on analyzing laser speckle patterns for precise measurements.
• Accurate quantification of speckle displacements is vital for detecting object faults and deformations.
• Existing methods face challenges in handling complex, multi-directional movements.

Purpose of the Study:

• To develop an advanced algorithm for accurate speckle displacement measurement.
• To integrate convolutional neural networks (CNNs) for optimizing grid overlap calculations.
• To enable simultaneous detection of multiple translational movements in object parts.

Main Methods:

• Image segmentation into overlapping grids.
• Fourier-based image registration for displacement quantification.
• Convolutional neural networks (CNNs) for optimizing grid overlap parameters.
• Monte Carlo simulation with grid search and k-fold cross-validation for hyperparameter tuning.

Main Results:

• The developed algorithm accurately quantifies speckle displacements.
• The CNN-based approach effectively optimizes critical overlap parameters.
• The method demonstrated capability in detecting multiple translational movements.
• Validation was successful using simulated and real speckle patterns.

Conclusions:

• The proposed method offers a precise and versatile approach for speckle displacement analysis.
• Integration of CNNs significantly enhances the accuracy and robustness of the algorithm.
• This technique holds promise for advanced quality control and deformation analysis in various applications.