Human pose estimation is the basis of human action recognition, and it is the guarantee to realize human-computer interaction. Occlusion, interference, complex background and different human scales have great influence on single-human pose estimation. To obtain more semantic information and reduce the impact of different human scales, an improved SNHRNet deep learning network is proposed to achieve more accurate single-human pose estimation. The improved SNHRNet network is established based on the HRNet network. The optimized NHRNet network is obtained by cutting the low-resolution high-level features of the HRNet. Compared with HRNet, the parameters of the NHRNet network are reduced, but the attention mechanism is missing. On this basis, the SE module is integrated into the Basic residual module to make up for the lack of attention mechanism in NHRNet network, and the SNHRNet network model comes into being. Finally, the public dataset is used for single-human pose estimation experiment. The results of relevant experiment show that the proposed SNHRNet network is effective for single-human pose estimation.