Video Knowledge Discovery Based on Convolutional Neural Network

Under the background of Internet+education, video course resources are becoming more and more abundant, at the same time, the Internet has a large number of not named or named non-standard courses video. It is increasingly important to identify courses name in these abundant video course teaching resources to improve learner efficiency. This study utilizes a deep neural network framework that incorporates a simple to implement transformation-invariant pooling operator (TI-pooling), after the audio and image information in course video is processed by the convolution layer and pooling layer of the model, the TI-pooling operator will further extract the features, so as to extract the most important information of course video, and we will identify the course name from the extracted course video information. The experimental results show that the accuracy of course name recognition obtained by taking image and audio as the input of CNN model is higher than that obtained by only image, only audio and only image and audio without ti-pooling operation.