Multiple Dark Soliton Tracking and Sorting System: Combining Machine Learning with Physics

In cold atom experiments, data comes in the form of images which often suffer information loss, inherent in the techniques used to prepare and measure the system. Moreover, the use of traditional fitting-based analyses on these images necessitates additional processing, which leads to further information loss. This is particularly problematic when the processes of interest are complicated, such as interactions among excitations in Bose-Einstein condensates (BECs). In our work, we aim to identify and track kink solitons in BECs, which are associated with local decrease in condensate density. While the traditional approach performed well in finding locations of single solitons, it was not generalizable to more complex cases. To overcome this limitation, we propose a framework combining machine learning and traditional analyses to detect, locate, and characterize multiple solitonic excitations. In particular, we developed a Physics-Informed Excitation (PIE) classifier, which sorts solitonic excitations into categories (e.g., kink solitons, solitonic vortices or canted solitons) based on correlations between fitting parameters from image segments. This combined framework can be used to detect multiple solitons, extract physical parameters, and fine-classify each solitonic feature.