Existing occluded pedestrian re-identification methods mainly utilize convolutional neural networks to realize the feature matching under different camera perspectives. Due to the complex occlusion situation, the accuracy of occluded pedestrian re-identification is not so satisfied where convolutional neural networks are utilized to extract local features. Convolutional neural network is unique in its ability to capture local features, but its global modeling ability is weak. In contrast, Vision Transformer (ViT) can efficiently extract global features from shallow layers with more spatial information and obtain intermediate features with high quality. To deal with the above issues, ViT is here introduced into the residual network to construct a dual-branch hybrid network of residual network and visual converter (DB-ResHViT), where the ViT branch is utilized to reduce training errors, while the residual-ViT branch is utilized to construct the global correlation of feature sequences extracted by the residual network. The proposed network proposes a novel data augmentation module, called partial image patch pre-convolution module (PPPC), which is utilized to input the extracted partial image patches into the pre-convolution network to replace the original image patches to achieve the goal of introducing local features into the ViT branch. In addition, the proposed network designs a novel module integrating residual and mobile vision transformer, called RMV Module, which is utilized to establish the global correlation of local features extracted by the residual network to achieve the goal of reducing the computational cost and improve the re-identification accuracy. Experimental results of a large number of occluded pedestrian re-identification datasets demonstrate that the performance of the proposed method is superior to other advanced methods. [ABSTRACT FROM AUTHOR]