In situ liquid STEM movie of the growth of gold nanoparticles side by side with its deep learning assisted denoised version
Résumé
With this video, we present the results of an innovative approach for denoising that combines artificial intelligence and in situ liquid scanning transmission electron microscopy experiments. Our approach enables the differentiation and extraction of material-derived signals from unwanted background fluctuations and noises in liquid dark-field STEM sequences using kinematic simulations. Besides considering the size and shape dispersions of nanoparticles, our simulations account for a significant obstacle in studying NP growth by liquid cell TEM: the formation of NPs on the opposite membrane of the cell that contribute to the random background fluctuations because they are imaged way out of focus. Considering these challenges, our method effectively denoises this video, elevating the signal-to-noise ratio from 1 to 7, above the threshold value of 5 set by the Rose criterion. Consequently, this work facilitates the study of nanoparticle growth mechanisms with improved statistics and fewer acquisition constraints. This will help to investigate surface adsportion on both gold nanocubes and nanorods within the context of nanoparticle synthesis.