Facial expression recognition has a very important application prospect in computer vision fields such as human-computer interaction and emotion calculation. Aiming at the challenges of complexity, diversity, occlusion and illumination of facial expression recognition, a new end-to-end network is proposed, and attention mechanism is applied to automatic expression recognition. The new network architecture consists of four parts: feature extraction module, attention module, reconstruction module and classification module. By extracting image texture information from LBP features, the tiny motion of face is captured and the network performance is improved. attention mechanisms can make neural networks pay more attention to useful features. we combine LBP features and attention mechanisms to improve the attention model to improve the recognition accuracy. applying the newly proposed method to two representative expression datasets, namely JAFFE、CK、FER2013 and. the experimental results show that the accuracy of facial expression recognition on three data sets is 98.95%,98.95% and 79.89%, respectively. It is proved that the method is beneficial to improve the recognition rate of facial expression and is advanced.