In order to solvethe problems existing in facial expression recognition,such as insufficient ability to obtain key information,low recognition rate and easy to overfit the model,a new facial expression recognition model(IERNet)was developed by using ResNet18 as the basic network.By introducing ECA attention mechanism,IERNet constructs two different kinds of attention residual units and forms an attention residual module,so as to enhance the ability to extract the key features of deep expressions.Iception module is introduced to extract the multi-scale shallow information of the image.By introducing the two modules at the same time,the robustness of the network is enhanced and the recognition rate of the model is improved.Finally,we use the global average pool combined with Dropout technology to replace the full-connection layer,which can effectively prevent the overfitting problem of the model and simplify the model. According to the experimental data,this paper has achieved good results in the public expression data sets CK+and FER2013,with the accuracy rate reaching 97.778%and 73.558%respectively.