METHOD FOR FORMATION OF DECISIONMAKING DEVICE THRESHOLD BASED ON NEUROREGULATOR Russian patent published in 2020 - IPC G06N3/02 

FIELD: data processing.

SUBSTANCE: invention relates to a method of forming threshold of object classification resolver. In the method, in the training mode when recording the change in the information parameters at the output of the parametriciser, which selects a set of secondary parameters from the set of input data, which are determining for the task being solved, as well as from the setting in the form of required binary responses, a training sample is formed and the artificial neural network (ANN) is trained as a neuroregulator, in which a threshold is formed, based on which a real output reaction Z is calculated in the resolver, and in the subtractor a control error signal is calculated, the neural network weight coefficients are corrected using the back propagation error algorithm, when training an ANN, a Jacobian of a decision device is used, which is a derivative of the Heaviside function; to calculate the Jacobian, a sigmoid function is used as an approximation of the Heaviside function; in the regulation mode, the trained neural regulator generates a threshold, and the parametriciser generates secondary parameters which are compared with the obtained threshold, after which a real output reaction Z of the resolver is generated in a binary form.

EFFECT: technical result consists in improvement of object classification accuracy.

1 cl, 2 dwg