METHOD FOR FORMING A TRAINING SAMPLE FOR CONTROL SYSTEMS OF UNMANNED ELECTRIC TRAINS Russian patent published in 2022 - IPC G06K9/62 G06V10/72 G06V10/82 G05D1/00 

FIELD: machine learning methods.

SUBSTANCE: invention relates to a machine learning method specially adapted for images. The method for forming a training sample for control systems of unmanned electric trains consists in receiving information from sensors scanning the surrounding space, preprocessing the information received and recording the initial set of frames, highlighting objects on the frames and classifying them. At the same time, a random subset of frames is selected in such a way that it contains objects of all classes in equal quantities, and the required number of all objects present in such a subset is checked. For each class of objects to be detected, the compliance of the empirical distribution of all variables with the required distribution law is checked, and in case of its inconsistency, the selected frames are filtered. Next, additional information is generated to further check the integrity of the sample, after which the frames of the sample are analyzed to calculate the values ​​of the class variables and the quality criteria of the obtained sample, such as completeness, representativeness, consistency, homogeneity and integrity, are evaluated, and if they match, the resulting sample is recorded, completing its formation.

EFFECT: improving the quality of formation of the training sample.

1 cl