Correct testing of the quality of convolutional networks of artificial neurons, taking into account the real conditions of their operation

Background. Currently, multilayer convolutional networks of artificial deep learning neurons are actively used to recognize people's faces. Their testing is carried out according to the ISO/IEC 19795-1-2007 standard by testing laboratories in unfriendly countries, which may distort the test results. Materials and methods. The basic international standard stipulates the volume of the test base of real people’s faces. It is possible to significantly reduce the size of the test base through morphing synthesis of new biometric images by crossing the images of parents according to the domestic standard GOST R 2633.2- 2010. At the same time, incorrect crossing of parent images can lead to a distortion of the test results. The situation is complicated by the fact that the neural network face recognition tool will work with real data of people's faces of different quality. Results. It is proposed to eliminate the threat of possible distortion of test results by providing the testing laboratory by the customer with a number of statistical points describing the real working databases of people’s faces. It is shown that in addition to mathematical expectation and standard deviation, it is necessary to use the third and fourth statistical moments. When calculating statistical points, it is proposed to train the tested neural network to recognize specific biometric images of the faces of people who have given their consent to the use of their personal data.