CNN-GRU-Transformer Human Activity Recognition Model Based on Feature Fusion

In recent years, human activity recognition technology has received great attention in various fields. The focus of this technology is to use wearable sensors connected to various locations on the human body to identify and classify human activity data. At present, many recognition algorithms can already recognize simple human activities. And achieved good results. The recognition effect of complex human activities collected by multiple sensors is poor. Therefore, this article proposes a CNN-GRU-Transformer human activity recognition model based on feature fusion. Fusion of data features from multiple sensors. Further explore the temporal dependence of the fused features. GRU can extract short-term temporal information from data. Transformers can effectively extract long-term dependencies from data. The combination of the two can effectively capture global features in the data. Improve the generalization of the model. The model was tested on the MSHA dataset and OCA dataset. The recognition accuracy reached 98% and 96% respectively. And compare and analyze with some existing models. This proves the superiority of the model proposed in this article.