Using Diffraction Deep Neural Networks for Indirect Phase Recovery Based on Zernike Polynomials

Area of Science:

• Optics and Photonics
• Artificial Intelligence in Imaging

Background:

• Traditional phase inversion techniques in imaging systems face challenges with sensor speed and system complexity.
• Accurate phase distribution information is crucial for monitoring and adjusting imaging system performance, particularly in adaptive optics.

Purpose of the Study:

• To propose and validate an indirect phase retrieval method using a diffraction neural network.
• To overcome the speed and complexity limitations of conventional phase inversion techniques.

Main Methods:

• An indirect phase retrieval approach utilizing a diffraction neural network with multiple diffraction layers.
• Reconstruction of Zernike polynomial coefficients from incident beams with distorted phases via non-source diffraction.
• Network training and simulation testing to evaluate performance for single-order and multi-order phase inversion.

Main Results:

• The trained network demonstrated capability for single-order phase recognition and multi-order composite phase inversion.
• Analysis confirmed the network's generalization and the impact of network depth on restoration accuracy.
• Achieved an average root mean square error of 0.086λ for phase inversion.

Conclusions:

• The proposed diffraction neural network offers an effective indirect phase retrieval method.
• This research provides new methodologies for phase recovery in adaptive optics systems.
• The approach shows promise for improving the performance and adaptability of optical imaging systems.