An Efficient Channel Attention CNN for Facial Expression Recognition

In order to overcome the difficulty of extracting facial expression features by neural network model, and the problems of complicated training process and parameter redundancy caused by stacking deep network structure, this paper proposes a CNN model that introduces an attention mechanism. In this paper, facial expression images are used as the object, and the research of cfacial expression recognition based on convolutional neural network is carried out. Based on the construction of natural expression feature views in the natural environment, the automatic data enhancement technology for deep convolutional neural networks is introduced, and the attention mechanism is combined to adjust the weights of different channels, and the facial expression recognition with texture information extraction as the traction is established deep learning model and facial expression recognition mechanism. At the same time, we propose a combined loss function. The effectiveness of this method is verified on the FER2013, FERplus, CK+, SFEW and RAF-DB data sets, and good results have been achieved.