近年来，随着深度学习方法在人脸微表情识别中的应用和发展，大规模人脸微表情数据集在识别准确率提升方面扮演的角色越来越重要。针对目前人脸微表情数据集样本少及标注困难等问题，本文提出了基于对抗神经网络的人脸动作单元自动生成方法，根据不同人脸动作单元及强度组合出多样化的连续微表情图像数据，实现大规模人脸表情数据集自动生成任务。

本文首先根据人脸动作编码系统的标准和数据实际情况建立了数据分类标准，并将满足条件的数据有序提取出来；其次，根据面部基准尺度确定了人脸在图片中的位置，实现了人脸的裁剪；最后，在StyleGAN框架下，实现了真实人脸的潜在代码生成方法；并使用潜在代码确定语义边界后，利用线性插值的方法生成了基于动作单元变化的连续人脸表情图像。

 In recent years, with the application and development of deep learning methods in facial micro expression recognition, large-scale facial micro expression data sets play an increasingly important role in improving recognition accuracy. In order to solve the problems of few samples and difficult labeling of facial micro expression data set, this paper proposes an automatic generation method of facial action unit based on counter neural network. According to different facial action units and intensity, a variety of continuous micro expression image data are combined to realize the task of automatic generation of large-scale facial expression data set. Firstly, according to the standard of face action coding system and the actual situation of data, the data classification standard is established, and the data satisfying the conditions are extracted orderly; Secondly, the position of the face in the image is determined according to the face reference scale, and the face clipping is realized; Finally, in the framework of stylegan, the potential code generation method of real face is implemented; After the semantic boundary is determined by the latent code, the linear interpolation method is used to generate the continuous facial expression image based on the change of action unit.