Automatic crack detection on road surfaces is an important task for supporting the quality control of road infrastructure in transportation. Various methods have been proposed for crack segmentation, but their accuracy is still limited. To improve the effectiveness of crack detection, we propose the Swin Transformer-V2 UNET for Crack Segmentation Network model (STUCNet) for crack recognition. The proposed model combines the advantages of the Swin Transformer-V2 into the encoding module of the UNET-based architecture to enhance the quality of semantic image segmentation. Specifically, the model integrates Swin TransformerV2 with shifted windows as the encoder to extract contextual features for crack segmentation. The symmetric decoder is based on a convolutional neural network with attention designed to perform up sampling operations to restore the spatial resolution of the feature map. We evaluate the STUCNet model on a large dataset containing cracks collected in different contexts. Compared to current advanced models, the proposed method achieves state-of-the-art (SOTA) results for crack segmentation.