Удосконалення технології розпізнавання облич за допомогою згорткових нейронних мереж

Abstract

The work is devoted to the study of methods of construction and algorithms of convolutional neural networks in the context of facial recognition to identify ways to improve existing models. The purpose of the study is to find and implement approaches to improve the face recognition system, which allows to recognize objects with increased recognition accuracy without slowing down. The object of research in this work is the process of face recognition using convolutional neural networks. The subject of research is the architecture and capabilities of convolutional neural networks to solve the problem of face recognition in two-dimensional images. The result are modifications of neural networks trained on various datasets, analysis and comparison of the obtained models on the parameters of recognition quality and learning time on CPU processors, as well as processors with accelerators – GPU and TPU. The scientific novelty is to obtain a more efficient architecture of the convolutional neural network, which improves the accuracy of face recognition. The practical importance is the accumulation of recommendations for the development and training of convolutional neural networks for use in face recognition systems based on the analysis of modern architectures and the principles of existing models.