Single shot imaging for cold atoms with machine learning

The ability to probe atoms in high spatiotemporal resolution give us a powerful tool investigate the quantum properties of ultracold gases. Plenty useful imaging methods, including absorption imaging, phase contrast and fluorescence have been implemented detect atoms. Among them, is most widely used method cold lab. However, traditional affected by perturbations such as interference between optical elements, fluctuation laser power, frequency, spatial position, resulting residual structured noise decrease quality. Especially for regions with lower density or longer time-of-flight, large number repetitions are often required obtain better signal-to-noise ratio, which would increase time cost induce other noise. One must two pulses suppress A charge coupled device (CCD) higher frame transfer rate like fast-kinetic mode be improve In this paper, single-shot atom machine learning proposed, only one required, corresponding background image can generated through neural network an autoencoder. This effectively reduces striped significantly improves quality, makes it possible imaged multiple times single cycle.