2D shape reconstruction of irregular particles with deep learning based on interferometric particle imaging.

Area of Science:

• Particle Imaging and Characterization
• Computational Fluid Dynamics
• Machine Learning Applications

Background:

• Interferometric Particle Imaging (IPI) is a key technology for particle measurement.
• Directly obtaining particle shape information from IPI is difficult due to complex speckle patterns.
• Irregular particle shapes pose a significant challenge for traditional IPI analysis.

Purpose of the Study:

• To develop a deep learning method for reconstructing irregular particle shapes from IPI data.
• To address the challenge of extracting shape information from interference-defocused speckle patterns.
• To improve the accuracy of particle shape analysis in IPI.

Main Methods:

• Implementation of a convolutional neural network (CNN) for image reconstruction.
• Utilizing sparse features of sand particles in the deep learning model.
• Introduction of the negative Pearson correlation coefficient as a specialized loss function.
• Validation of the method using experimental IPI data.

Main Results:

• Successful reconstruction of defocused images of sand particles.
• Demonstrated effectiveness of the CNN in capturing particle shape characteristics.
• The proposed network structure showed superior performance compared to other CNN architectures.
• Accurate shape reconstruction of irregular particles was achieved.

Conclusions:

• The developed deep learning approach effectively reconstructs particle shapes from IPI data.
• The CNN method offers a viable solution for overcoming limitations in IPI shape analysis.
• This technique enhances the capability of IPI for detailed particle characterization.