Research on Expression Recognition Algorithm Based on Attention Mechanism

Facial expressions carry the emotional state of human beings and are one most powerful signals for humans to express themselves. Therefore, research facial expression recognition has very important significance broad application prospects. However, in complex scenes, blurring images, interference redundant information, similarity will cause unsatisfactory recognition. In response these shortcomings, this article adds a convolutional attention module basic Resnet network structure, infers map turn along two independent dimensions (channel space), performs adaptive feature optimization, captures reduces faces. The information; introduction island loss ISlandloss classification optimize distance between classes, so that can learn more discriminative features, improve discrimination expressions. Compared with existing mainstream algorithms, method certain advantages objective evaluation indicators. experimental results on FER2013 show accuracy algorithm is 73.85%, which higher than 71.23% network.