Distorted pulse detection integrating two-photon absorption and an improved convolutional neural network with self-attention

Le résumé fourni par la source

A distorted pulse detection method for ultrafast laser combining two-photon absorption (TPA) with an improved convolutional neural network (CNN) with self-attention mechanism is presented in this work. The dataset was constructed by mixing simulated and experimentally acquired autocorrelation signals from the three pulse types(normal pulses, pulses with cubic spectral phase, and double pulses). A lightweight 1D CNN and an improved CNN with self-attention were developed with hyperparameters optimized using Bayesian methods. The two networks achieved the overall accuracy about 96% and 97% on the test set, with common recognition rates for double pulses over 99%. It indicates that CNN demonstrates a strong capability in resolving and classifying pulse autocorrelation signals. Moreover, for the highly similar autocorrelation traces of normal pulses and pulses with cubic spectral phase, the self-attention mechanism enables the detection of subtle differences between these patterns. This research advances data-driven laser optimization and maintenance.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.