Multiclass Live Streaming Video Quality Classification Based on Convolutional Neural Networks

Automatic Control and Computer Sciences 2022
Tianhua Chen, Elans Grabs, Ernests Pētersons, Dmitry Efrosinin, Aleksandrs Ipatovs, Nikolajs Bogdanovs, Dmitrijs Rjazanovs

E-sports live streaming video is rapidly coming into people’s lives. High-quality video is an essential factor affecting users’ perception. This paper presents conventional network traffic analysis methods for traffic intensity selection as a feature combined with deep learning classifiers for streaming videos classification with different resolutions and frame rates per second. According to the experimental results, the convolution neural networks showed the best results in multiclass classification with accuracy as high as 97%. This superiority can help E-sports operators to improve the quality of live streaming videos and provide differentiated services for their users. Furthermore, the article describes research on the performance of various deep learning classifiers with different hyperparameters. The number of filters in convolution layers and training batch size can significantly affect classification performance according to testing results. It is still necessary to avoid hyperparameters’ designated values significantly influencing the classification results.

Keywords
multiclass classification, traffic intensity, convolutional neural networks, deep learning, features selection, hyperparameters evaluation
DOI
10.3103/S0146411622050029
Hyperlink
https://link.springer.com/article/10.3103/S0146411622050029

Chen, T., Grabs, E., Pētersons, E., Efrosinin, D., Ipatovs, A., Bogdanovs, N., Rjazanovs, D. Multiclass Live Streaming Video Quality Classification Based on Convolutional Neural Networks. Automatic Control and Computer Sciences, 2022, Vol. 54, No. 5, pp.455-466. ISSN 0146-4116. e-ISSN 1558-108X. Available from: doi:10.3103/S0146411622050029

Publication language
English (en)