In order to solve the problem of low detection degree and high missed detection rate of occluded pedestrians in pedestrian detection, an attention mechanism based UAST-RCNN network is proposed, which is improved on the basis of Faster-RCNN network. Firstly, Swin-Transformer is selected as the backbone network to improve the global receptive field by using a window multi-head selfattention mechanism. Then, the feature pyramid is improved for the quality of feature samples through the hierarchical resampling module, and a progressive focus loss function is introduced to balance positive and negative samples. Finally, in the preprocessing stage of the experiment, the improved data preprocessing was used to extend the City Persons dataset for multi-scale training. The experimental results show that the algorithm has a significant improvement in the detection of occluded pedestrians compared with the original model, in which the AP is increased by 6. 3%, and the MR (miss rate) is decreased by 4. 1%. The feasibility of the proposed algorithm in pedestrian detection is verified, and it can meet the detection requirements of occluded pedestrian scene.