Анализ алгоритмов поиска объектов на изображениях с использованием различных модификаций сверточных нейронных сетей

Александр Анатольевич Сирота; Елена Юрьевна Митрофанова; Анастасия Ивановна Милованова

doi:10.17308/sait.2019.3/1313

Authors

Александр Анатольевич Сирота Voronezh State University
Елена Юрьевна Митрофанова Voronezh State University
Анастасия Ивановна Милованова Voronezh State University

DOI:

https://doi.org/10.17308/sait.2019.3/1313

Keywords:

object detection, object classification, neural networks, deep learning, RCNN, Fast RCNN, Faster RCNN

Abstract

The article deals with the tasks of classifying and detecting objects in images using computer vision algorithms. The description of the main types of algorithms and methods for searching objects based on the use of deep neural networks is given. A comparative analysis and modeling of neural network algorithms for solving the problem of classification and search for objects in images has been carried out. The results of testing neural network models with different architectures on VOC2007 and COCO data sets are given. The results of the study of recognition accuracy are analyzed depending on various learning hyper-parameters. The change in the time value of determining the location of an object depending on various neural network architectures is investigated.

Author Biographies

Александр Анатольевич Сирота, Voronezh State University

prof., Head of the Chair of Information Processing and Security Technologies at Voronezh State University
Елена Юрьевна Митрофанова, Voronezh State University

docent at the Chair of Information Processing and Security Technologies at Voronezh State University
Анастасия Ивановна Милованова, Voronezh State University

Master of 2 year training program «Information Systems Security», Computer Sciences Faculty, Voronezh State University