Autonomous dark soliton detection

Solitary waves (solitons) are non-dispersing localized traveling waves that retain their size, shape, and speed as they move, and even when they collide with one another. In a repulsively interacting 1D Bose-Einstein condensates (BEC), the soliton velocity governs the width, depth, and even stability of a soliton. To study solitons dynamics, a number of absorption images need to be manually analyzed to obtain the number of solitons present and their position. These solitons must be distinguished from a background of additional excitations. ~ We developed an automated two-step protocol for detecting dark solitons in two-component BECs. Our algorithm combines two neural networks pre-trained using simulated and real data to: (1) identify the number of solitons present in BEC and (2) determine their position. This automated detector highlights the applicability of machine learning-driven feature detection, rather than traditional curve fitting, to streamline cold atom research.