Facial Expression Recognition Method Based on Multi-scale Efficient Channel Attention Mechanism

Facial expression recognition is an important research content in the field of computer vision. Aiming at the problems of incomplete feature extraction and weak pertinence of important features in the process of facial expression recognition based on traditional convolutional neural networks, this paper proposes a facial expression recognition method based on multi-scale efficient channel attention mechanism. Firstly, multi-scale feature extraction layers are used to extract more abundant expression feature information; Secondly, the efficient channel attention mechanism is introduced into the multi-scale feature extraction layer to improve the response of the model to important features; Finally, the shallow layer feature information and the deep layer feature information are fused by cross-connection to avoid the loss of shallow layer edge feature information when the network layers are deep. A method to balance the number of different categories of expressions is proposed to reduce the large difference between different types of expressions. The results of the proposed method are comparable to or better than many classical algorithms.