A machine learning model has been developed to search events of production and decay of a hypertriton in nuclear emulsion data, which is used for measuring the binding energy of the hypertriton at the best precision. The developed model employs an established technique for object detection and is trained with surrogate images generated by Monte Carlo simulations and image transfer techniques. The first hypertriton event has already been detected with the developed method only with 10−4 of the total emulsion data. It implies that a sufficient number of hypertriton events will soon be detected for the precise measurement of the hypertriton binding energy.