Deep Learning Method for Image Processing in Cold Atom Experiments

Image processing is a fundamental part of cold atom experiments. Many such experiments use absorption imaging, which requires fitting the data to some distribution to extract valuable experimental metrics. Traditionally this is done using least squares fitting algorithms, however they are highly sensitive to noise, computationally costly, and rely heavily on the accuracy of the initial guess. We present a deep learning method that directly processes raw absorption images to output Gaussian fit parameters. By leveraging convolutional neural networks, we achieve greater robustness and speed compared to a traditional fitting algorithm.