A method and apparatus for analysing a deep learning (DNN) based classroom learning behaviour. The method comprises: first, recalculating a transparency estimation value by measuring the credibility of a foreground and background pixel pair so as to obtain a first transparency mask of a first image; then, generating a new picture by superposing greyscale information, obtaining a second transparency mask of the first image, and further modifying the first transparency mask of the first image; and finally, using the modified first transparency mask to extract a foreground target of a certain frame of image in a video, and further extracting a background target in the first image, and performing facial expression analysis on the extracted foreground target and background target by means of a deep neural network. By means of the method, a new classroom behaviour analysis solution can be provided by comprehensively utilising the credibility and greyscale information of a foreground and background pixel pair in a certain frame of image in a video.
展开▼