Practical and Accurate Evaluation of Numerical Aperture and Beam Quality Factor in Photonic Crystal Fibers by Mechanical Learning

This paper presents a convolutional neural network (CNN) model, enhanced with the convolutional block attention module (CBAM), designed to accurately predict the beam quality factor M 2 , and numerical aperture (NA) of photonic crystal fibers. The integration of CBAM significantly improves the model's feature extraction capability by enabling it to focus on key features and filter out irrelevant information. Simulation results demonstrate that the model achieves a mean relative error of only 0.381% for M 2 and 2.293% for NA, outperforming convolutional models without attention mechanisms. With a prediction time of approximately 7 ms, the model allows for rapid and efficient predictions of M 2 and NA. Moreover, when the noise factor remains below 0.32, the model's prediction error shows minimal fluctuation, highlighting its robustness. Comparative experimental analysis further validates the model's effectiveness. This approach offers a reliable and efficient solution for fast, accurate measurement of M² and NA, with significant implications for the prediction and analysis of beam performance in various applications.