Rapid Diversion of Remote Audio and Video Data in QoS Weak Network Environment

In order to effectively manage audio and video data and extract important information from videos, audio and video automatic classification technology has become the main way to solve this problem. Audio and video data consists of image frames, and the relationship structure of their internal information is relatively complex. Traditional manually designed features cannot effectively represent the complex information in video data. The use of deep learning technology can extract relatively complete feature information and improve the accuracy of video classification. Therefore, this article studies a video classification method based on deep learning in a QoS weak network environment. This paper proposes a multi-channel convolutional network (MCCN) video classification method based on depth metric learning to address the impact of semantic changes in video on classification results and how to improve intra class similarity and inter class dispersion in the video classification process. Based on the network model of feature fusion at different scales, a multi-channel convolutional video classful network is designed. In order to enable the network to learn intra class similarity and inter class dispersion, an interval allocation function based on negative sample pair semantic distance is proposed in the metric learning structure, which makes the network pay more attention to difficult to distinguish samples and perform both metric learning and classification tasks during training. The experimental results show that this method can improve the accuracy of video classification.