Micro-expression refers to the expression that people spontaneously generate when they want to hide their true emotions. The duration is within 1/5 s. Their facial muscles are short and small in amplitude. The limited data set makes feature extraction difficult. These factors bring huge challenges to its recognition. In response to these problems, this paper proposes a recognition method based on image preprocessing technology and dual-branch network. Firstly, an advanced face alignment network is used to obtain effective facial regions, and then Euler video magnification technology is used to capture the subtle changes in facial movements. Optical flow information is used as the feature of the video sequence, and the feature maps of different combinations of optical flow information are input into the two-branch classification network to obtain the expression label output. Experiments aere performed on the two public data sets of SMIC and CASME Ⅱ, and the leave-one-subject-out(LOSO) is employed for cross-validation. The accuracy can reach 0.545 and 0.584, respectively. The quantitative and qualitative analysis of the experimental results verify the effectiveness of the proposed method.