Mapping different soliton state in the automatic spatiotemporal mode-locked fiber laser based on convolutional neural network

A novel approach combining deep learning (DL) model with traditional mode-locked techniques is proposed to achieve the dissipative solitons (DSs) and noise-like pulses (NLPs) automatic spatiotemporal mode-locked fiber laser (MLFL). By utilizing the convolutional neural network (CNN), mapping between the rotation angle of the electric polarization controller (EPC), the output spectrum with different states of solitons and the complex beam profile is first established. Analysis of the training and testing loss curves demonstrates that the DL model effectively learns the latent spatial-spectral features necessary for accurate prediction. When supplied with random beam profiles, predicting the corresponding spectral characteristics and the rotation angle of the EPC almost perfectly matches with the experimental results. The proposed DL model enables the accurate reproduction of the rotation angle of the EPC and output spectrum with different types of soliton, and facilitating the automatic tuning of EPC can enhance stable mode-locked state in the experiment, thereby promoting the progress of the automatic fiber laser system.