Prediction of the whispering-gallery modes in spherical hyperbolic metamaterial cavity based on deep learning

Microcavity structures with whispering-gallery modes (WGMs) have significant applications in developing advanced optical devices, making them a cornerstone in the field of optics. However, the identification of high-order WGMs remains challenging due to their smaller mode volume and higher optical field density. To address this issue, we construct a dataset of WGMs consisting of 1869 images and evaluate its performance using the YOLO algorithm. Furthermore, architectural modifications are introduced to enhance the algorithm's prediction accuracy, achieving an 8.6 % increase in mAP compared to the baseline. The improved model demonstrates reliable performance in predicting the order of WGMs, providing a valuable approach to solving the recognition challenges associated with complex optical modes.