METHOD OF OBJECTS RECOGNITION USING NEURAL NETWORKS Russian patent published in 2019 - IPC G06K9/62 G06N3/08 

FIELD: data processing.

SUBSTANCE: invention relates to object recognition. Obtaining an image of an object using a camera comprising an image frame processing module using one or a plurality of multiple neural networks F, detecting using said image module and selecting image regions O_i with objects of interest, transferring image regions O_i with selected objects of interest at input of one or plurality of several neural networks F inside chamber, calculating, along selected areas, significant features of objects f(O_i), transmitting same image regions O_i with selected areas of interest to third-party more powerful server, on which the second one or more complex neural networks G is installed, significant features of objects g(O_i) on external neural network G, transmitting back to camera features of objects g(O_i), performing neural network re-education in chamber F using images accumulated in the camera O_i and values of feature vectors g(i) calculated using an external neural network G.

EFFECT: high speed and reliability of identifying objects with a camera by training a neural network built into the camera using a more powerful external neural network.

5 cl, 3 dwg, 5 tbl