A System for Image Classification Using Deep Convolutional Neural Networks

Abstract

A system of classification of color images by deep convolutional neural networks has been developed, which includes pre-processing of images, augmentation, construction and training of convolutional models, evaluation and selection of models based on multiple indicators, forecasting. A comparison of common and modern ResNet, EfficientNet, VGG, Inception, MobileNet architectures was carried out in terms of their learning and execution time, flops, accuracy, precision, recall and f1-score indicators on the selected data set. The robustness of the models to noise in images was investigated.