An improved HTNet micro-expression recognition method

Micro-expressions refer to brief and subtle changes in facial expressions, which correspond to different emotional states. Therefore, research on micro expressions is of great significance. At present, most of the micro expression recognition methods related to neural networks have good recognition performance, but the computational complexity needs to be improved. Therefore, this article proposes a lightweight Hierarchical Transformer Netwok (HTNet) model (LHTNet), which uses HTNet as the teacher model and the lightweight HTNet model as the student model to learn the feature representation of the teacher model. The feature distillation technique is used to extract the advanced features in the teacher model and transfer the feature knowledge to the student model learning to improve the performance of the student model. The deep separable convolution is introduced into the feedforward fully connected network (FNN) and the multi-head self-attention mechanism in the transformer layer to replace the standard convolution layer. Experiments on CASME2, SAMM and SMIC datasets show that the improved model has achieved significant performance improvement in micro-expression recognition tasks, and the number of model parameters and computational complexity have also been effectively reduced.