Dehazing is a crucial task in computer vision that aims to generate high-resolution images from images affected by haze or fog. We propose a novel dehazing network, DehazeSwinUnet (DSUnet), which is an improved version of the Swin Transformer designed to enhance the capability of detail recovery and dehazing quality. To overcome the shortcomings of existing methods in capturing fine details and global features, DSUnet introduces a DSUnet Block with integrating parallel convolution and window-based multi-head self-attention (W-MSA). In addition, we propose a new normalization method, RegionNorm, and a new window partitioning approach, Dynamic Window Partition. Furthermore, the study incorporates two innovative modules: the Channel-Spatial Branch Attention Module (CSBAM) and the Conditional Depth-Aware Reconstruction Module (CDARM), which are used to enhance feature representation and detail reconstruction, respectively. These contributions further improve the dehazing performance and quality of the generated images. The experimental results indicate that DSUnet achieves strong PSNR and SSIM scores on multiple benchmark datasets. Specifically, on the RESIDE indoor dataset, DSUnet achieves a PSNR of 43.77 and an SSIM of 0.998, validating its superiority and robustness.