Face Expression Recognition Combining Improved DeeplabV3+ and Migration Learning

Abstract

The main problem with current large-scale face expression recognition is uncertainty, which arises from ambiguous facial expressions, low-quality facial images, and the subjectivity of the annotator. To this end, an improved network model based on DeeplabV3+ is proposed in this paper. The model adopts the lightweight MobilenetV2 module as the backbone feature extraction network, which reduces the network model parameters and improves the semantic segmentation efficiency. The Convolutional Block Attention Mechanism module (CBAM) introduced in the encoder and decoder can improve the model feature extraction capability, realize the effective fusion of feature information of different layers, and improve the model segmentation accuracy. The training method of migration learning used to overcome the variability of sample distribution can enhance the generalization ability of the model, and adapt to the expression recognition tasks in different application scenarios. Finally, experimental validation was carried out on the RAF-DB and FER-2013 datasets. The results show that the facial expression recognition rate of the improved model in this paper reaches 88.90% and 75.61%, respectively, indicating the effectiveness of the method in facial expression recognition.