Research on Ensemble Convolutional Neural Networks in Digital Image Course experiments

Abstract

In various digital image processing courses, the accuracy of traditional expression recognition algorithms cannot meet practical needs. This paper studies facial expression recognition using a convolutional neural network and ensemble learning. We develop three subnetwork models based on the classical convolutional neural network, using the BN layer to increase the network convergence rate and regularization is used to prevent subnetwork overfitting. The optimization idea used in this paper is applicable to most experimental improvements and raises the standard of the experiments.