Research on Deep Learning Expression Recognition Algorithm Based on Multi-Model Fusion

As an important form of human emotion expression, expression has broad application prospects in the fields of human-computer interaction and emotion calculation. Aiming at the fact that the facial expressions in reality are affected by illumination and attitude angle, and the traditional expression extraction method can't effectively extract the features of strong expressive power, a deep learning expression recognition algorithm model based on multi-model fusion is proposed. The residual feature net-work is used to extract the expression features. Secondly, the feature weights of different feature maps are learned by embeding the squeeze and excitation network module in the residual network unit, and the feature re-calibration method is used to improve the network expression capability. Experiments show that this method has achieved good recognition results in the KDEF database.