Recovery of ultrashort optical pulses from 2D SHG-FROG traces based on modified Resnet-50 model

In this paper, we modify Resnet-50 deep-learning network to recover ultrashort optical pulses from their 2D SHG-FROG traces. Different from those reported methods, our model doesn’t need information of the true pulse, but evaluate accuracy of the recovered pulse by regenerating the 2D SHG-FROG trace and measuring its difference from the original FROG trace. The effectiveness of the proposed method is verified by recovering several types of pulses. Results demonstrate that the minimum of correlation coefficients between the regenerated and the original FROG traces can exceed 0.9503 even for the complex pulse with fluctuations.