Distortion correction for particle image velocimetry using multiple-input deep convolutional neural network and...

Area of Science:

• Optical Engineering
• Fluid Dynamics
• Machine Learning

Background:

• Aberrations in imaging systems degrade quantitative measurements like particle image velocimetry (PIV).
• Conventional adaptive optics systems using actuators have limitations in technical specifications and performance.
• Accurate flow measurements are crucial in various scientific and engineering fields.

Purpose of the Study:

• To develop an actuator-free adaptive optics method for correcting time-varying wavefront distortions.
• To enhance the accuracy of imaging-based flow velocimetry, particularly through challenging interfaces.
• To demonstrate a deep learning approach for real-time aberration correction.

Main Methods:

• A multiple-input deep convolutional neural network was designed, incorporating wavefront sensor data.
• An experimental setup with a deformable mirror was used to generate a dataset for training the neural network.
• The method was applied to imaging flow velocimetry across a fluctuating air-water interface.

Main Results:

• The deep learning model significantly improved image quality by correcting distortions.
• Flow measurement errors caused by the phase boundary fluctuations were reduced by 82%.
• The proposed technique demonstrated superior performance compared to traditional methods in specific scenarios.

Conclusions:

• The proposed actuator-free deep learning method effectively corrects time-varying wavefront distortions.
• This technique offers a promising alternative to conventional adaptive optical systems with insufficient actuator performance.
• The approach has the potential to advance quantitative imaging in complex and dynamic environments.