Accurate detection of small particles in digital holography using fully convolutional networks

Area of Science:

• Optical Engineering
• Particle Image Velocimetry (PIV)
• Combustion Science

Background:

• Particle detection is crucial for characterizing particle fields using digital holography.
• Background noise in digital holography can lead to spurious particles or loss of real particles, especially small ones.
• Accurate detection of small particles is a significant challenge in energy and combustion research.

Purpose of the Study:

• To develop an accurate method for detecting small opaque particles in extended focus digital holography images.
• To address the challenges posed by background noise in small particle detection.
• To improve particle field characterization for energy and combustion applications.

Main Methods:

• A deep learning approach utilizing modified fully convolutional networks (FCNs) was employed.
• The method focuses on detecting small opaque particles, such as coal particles.
• The model was trained and tested on extended focus images obtained from digital holography.

Main Results:

• The proposed deep learning model demonstrated good accuracy in detecting small opaque particles.
• Experimental validation confirmed the effectiveness of the method in challenging noisy conditions.
• The approach successfully mitigates issues of spurious particle generation and particle loss.

Conclusions:

• Modified fully convolutional networks offer a robust solution for small particle detection in digital holography.
• The developed method enhances the accuracy of particle field characterization in energy and combustion studies.
• This deep learning approach provides a reliable tool for analyzing small particles in complex environments.