A Novel Attention-based Neural Network for Video Scene Classification in Complex Background

As the video surveillance develops, research on video scene classification has increased accordingly. In order to further improve the accuracy of scene classification for intelligent surveillance video, a novel attention-based neural network was proposed. Firstly, given a video sequence, we used CNN (Convolutional Neural Network) to extract features of each its frames and later forward it to an LSTM (Long-Short Term Memory Network) to further obtain the temporal information. The LSTM outputs were then connected as the feature vector of video sequences. Finally, the feature vectors were forward to the attention part for weight calculation. The experimental results showed that our algorithm improved the video scene classification accuracy effectively.