通过面部表情来识别情绪已经广泛应用于日常生活中。在深度学习算法中，卷积神经网络(CNN)在面部表情识 别领域取得了巨大成功，但仍旧面临信息冗余和数据偏差的问题，影响着面部表情识别算法性能。因此提出了一种基于从面 部图像中提取的地标图的多尺度图卷积网络(GCN)。在 CK+、JAFFE、FER2013、RAF-DB 数据集上对所提出方法进行了仿 真。结果表明，所提方法优于AUDN、BDBN、SCNN等传统的深度学习框架，在不同数据集上准确率更高。

Recognizing emotions through facial expressions has been widely applied in normal life. Among deep learning algo‐ rithms, Convolutional Neural Networks (CNN) have achieved great success in the field of facial expression recognition. However, it still faces the problem of information redundancy and data deviation, which affects the performance of facial expression recognition algorithm. Therefore, a Multi-scale Image Convolutional Network (GCN) based on landmarks extracted from facial images is pro‐ posed. The proposed method is simulated on CK+, JAFFE, FER2013 and RAF-DB datasets. The results show that the proposed method is superior to AUDN, BDBN, SCNN and other traditional deep learning frameworks, and has higher accuracy on different data sets.