Facial Expression Recognition Based on Attention Mechanism

At present, traditional facial expression recognition methods of convolutional neural networks are based on local ideas for feature expression, which results in the model’s low efficiency capturing dependence between long-range pixels, leading to poor performance recognition. In order solve above problems, this paper combines a self-attention mechanism with residual network and proposes new model global operation idea. This first introduces basis finds relative importance location by calculating weighted average all then channel attention learn different features domain, generates focus interactive channels so that robustness can be improved; finally, it merges increase ability extract globally important features. The accuracy CK+ FER2013 datasets is 97.89% 74.15%, respectively, fully confirmed effectiveness superiority extracting