Deep learning enabled superfast and accurate M² evaluation for fiber beams.

Area of Science:

• Optics and Photonics
• Machine Learning Applications
• Laser Beam Characterization

Background:

• Accurate measurement of the beam propagation factor (M²) is crucial for laser beam quality assessment.
• Traditional M² measurement methods can be complex and time-consuming, requiring multiple beam profiles.
• Few-mode fibers present unique challenges for beam propagation analysis.

Purpose of the Study:

• To introduce a novel deep learning technique for predicting the M² of laser beams from few-mode fibers.
• To develop a fast and accurate method for M² evaluation using single beam patterns.
• To demonstrate the robustness and experimental feasibility of the proposed deep learning approach.

Main Methods:

• A deep convolutional neural network (CNN) was trained using simulated near-field beam patterns and their corresponding M² values.
• The CNN learned a mapping function to predict M² directly from single beam pattern images.
• The model was tested with simulated data, including noisy inputs and up to 10 eigenmodes.

Main Results:

• The deep learning model achieved an average prediction error of less than 2% for simulated few-mode fiber beams.
• Even with significant noise, the prediction error remained below 2.5%, demonstrating robustness.
• The M² estimation process takes approximately 5 ms per pattern, enabling real-time applications.

Conclusions:

• Deep learning offers a superfast and accurate method for M² evaluation with reduced experimental effort.
• The proposed technique is feasible for real-time M² determination using a single camera.
• The method is adaptable to other beam types with sufficient training data.