Human Activity Recognition (HAR) is widely used in fields such as monitoring and elderly care. Radar, as a non-contact sensor, has attracted attention due to its privacy protection and resistance to environmental factors such as lighting interference. Deep learning (DL) models have shown great potential in feature extraction from radar data. Due to the high cost of radar data, it is crucial to design a method that can extract more features on a small amount of data. This article proposes a parallel architecture SCNN-TRNN that combines convolutional neural networks (CNN) and recurrent neural networks (RNN), which combines spatial and temporal attention mechanisms for extracting spatial and temporal features from spectrograms. The experiment showed that the method achieved a classification accuracy of 97.50% on a radar dataset containing six daily activities.