CONVOLUTIONAL NEURAL NETWORKS FOR IMAGE RECOGNITION

Abstract

The purpose of the work, the results of which are presented within the framework of the article, was to study modern architectures of convolutional neural networks for image recognition. The article considers such architectures as AlexNet, ZFnet, VGGNet, GoogleNet, ResNet. A characteristic of the quality of image recognition for a neural network is the top-5 error. Based on the results obtained, it was revealed that at the moment the network with the most accurate result is the ResNet convolutional network with an accuracy rate of 3.57%. The advantage of this study is that this article gives a brief description of the convolutional neural network, and also gives an idea of the modern architectures of convolutional networks, their structure and quality indicators.