Human action recognition using Wi - Fi signals can significantly facilitate the development of human-computer interaction applications in a wide range of applications. A key challenge of existing research is constructing human action recognition models that can maintain stable performance under dynamic spatio-temporal conditions. In this paper, we propose an unsupervised adversarial domain adaptation model WiADG Pro, which can effectively capture the temporal features in the data. Experimental results on Widar 3.0 show that our proposed model can greatly improve the performance of the model in the domain adaptation experimental phase. And due to the unequal amount of data in the two domains, we also explore in depth the effects of different data amounts in the source and target domains and actionindependent domain information on the model performance.