Ultrashort pulse characterization and measurement is critical in the field of ultrafast and nonlinear optics. Here we present a method to reconstruct the complex pulse profile using a colinear frequency resolved optical gating (CFROG) acquisition combined with a convolutional neural network (CNN). The CFROG approach can be implemented with nonlinear nanoprobes for probing complex ultrafast optical fields. Typically, a CFROG trace is filtered and converted to a standard FROG trace which can then be processed by using the FROG retrieval algorithm to reconstruct both the amplitude and the phase profiles of the pulse. In this method, however, the reconstruction is often dependent on the subjective filtering step. In our approach, a CNN is trained with simulated unfiltered CFROG traces. Furthermore, we customize the CNN architecture to mitigate the ambiguity in the solution space and minimizes the error between the predicted and the input
|