The use of radar for human activity recognition has the characteristics of not being affected by external factors such as illumination and protecting user privacy, and millimeter wave radar as an important sensor has been widely used in attitude recognition. Based on this, a Doppler and Range decision-level Convergence Network model (DRCNet) is proposed in this paper. Firstly, the common human activity data is collected by millimeter wave radar, the doppler-time plot and range-time plot are obtained by preprocessing the original data. Then the convolutional neural network is built according to the characteristics of the generated human activity data set, and finally the final recognition result is output by decision-level convergence. The experimental results show that the recognition accuracy of DRCNet is improved compared with other models. The collected human activity can be effectively identified, and the recognition accuracy reaches 93.00%.