Modeling and Classification of Optical Beam Profiles Using Vision Transformer

A vision transformer (ViT) is developed to perform beam profile classification on beam profiles coupled out from silicon photonics (SiPh) gratings. The classification task is aimed to distinguish ‘focused’ and ‘sparse’ beam profiles, and the regions where the corresponding beam profiles are located above the SiPh gratings. Upon training with 1247 beam profile, the ViT model is able to perform 6-category classification task on 832 beam profiles with classification accuracy of 0.978. Since the training of ViT is probabilistic in nature, the ViT training is repeated for 200 times to test its robustness. Classification accuracy ranges from 0.952 to 0.981 is obtained.