| Autor: |
Alekseev, Viktor, Sidorenko, Valentina, Khusenov, Dodokhon, Andreyev, Andrey |
| Předmět: |
|
| Zdroj: |
AIP Conference Proceedings; 2023, Vol. 2476 Issue 1, p1-8, 8p |
| Abstrakt: |
Face recognition is a very interesting direction in terms of the emergence of behaviour and object behaviour. The automated control system with a distributed service personnel service system is identified by the access model and the possibility of performing routine maintenance on the equipment of various vehicle traffic control systems is provided. The use of imported biometric recognition applications in critical transportation systems is prohibited. In this regard, it is necessary to use domestic applications embedded in the access model. Time, coordinates of the position of employees and their biometrics (face recognition) are the basis for the implementation of the model of access of subjects to access objects. The paper proposes an approach aimed at solving the main disadvantages of convolutional neural networks - incomplete invariance of the algorithm for the degree of image brightness, noise and face location in the image. The proposed model includes the formation of a model using the structure of the Kolmogorov-Gabor series, which shows a greater accuracy of determination in comparison with the classical algorithm of convolutional networks. The level of recognition errors decreased by 1.2 ... 2.5 times, depending on the configuration of the perceptron and the set of images. The paper proposes the implementation of the model using full connectivity on each row of the perceptron, as well as criteria-based selection of test data models that do not intersect with the trained data sequence, which implements the principle of external complement. The proposed model connects a large number of features based on a structure that allows you to increase the glued fragments, based on Cartesian products and increases the degree in candidate models. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
