Fast modal analysis for Hermite-Gaussian beams via deep learning.

Area of Science:

• Optics and Photonics
• Machine Learning Applications
• Laser Physics

Background:

• Hermite-Gaussian (HG) beams are fundamental in laser physics, forming a complete basis for describing light modes.
• Accurate mode decomposition (MD) is crucial for laser applications but can be computationally intensive.
• Existing MD methods may lack speed, robustness, or the ability to extract both amplitude and phase information.

Purpose of the Study:

• To introduce a novel deep learning-based mode decomposition (MD) scheme for Hermite-Gaussian (HG) beams.
• To demonstrate the speed, accuracy, and robustness of this AI-driven MD approach.
• To enable fast acquisition of HG beam amplitude and phase information from single intensity images.

Main Methods:

• Training a convolutional neural network (CNN) using a large dataset of simulated HG beam patterns.
• Employing the trained CNN to perform MD on unseen simulated HG beam intensity images.
• Evaluating the scheme's performance with varying numbers of HG modes and in the presence of noise.

Main Results:

• Achieved an average prediction error of 0.013 for six HG modes.
• Demonstrated real-time MD capability with a processing time of approximately 23 ms per beam pattern.
• Showed robustness against noise, with prediction errors below 0.037 for heavily corrupted patterns.

Conclusions:

• The developed deep learning scheme provides a fast, economical, and robust method for HG beam MD.
• This technique allows for the retrieval of both mode amplitude and phase from single intensity images.
• The method has significant potential for applications in beam shaping, quality evaluation, resonator studies, and adaptive optics.