Water body extraction is a fundamental task in remote sensing image analysis. While existing approaches, such as traditional water index-based methods and deep learning models, perform adequately in general scenarios, they often struggle with complex water body boundaries and small-scale targets, resulting in blurred contours and inaccurate identification. To address these challenges, we propose a novel multi-branch parallel network for water body extraction, called WEFNet. The proposed WEFNet adopts ResNet34 as the backbone for feature extraction and introduces parallel Swin Transformer branches to enhance global contextual representation through hierarchical self-attention mechanisms. Moreover, we designed a Multi-scale Feature Fusion Module (MFFM) to aggregate features with different semantic information, thereby enhancing the model’s understanding of water body characteristics at different scales. To further improve boundary localization, we introduce an Edge Extraction Module (EEM), which utilizes multiple instances of Pixel Difference Convolution (PDC) to extract fine-grained edge features and integrates spatial attention to mitigate irrelevant noise. We conducted extensive experiments on the LoveDA and MSRWD datasets, achieving IoU scores of 0.7950 and 0.9598, respectively. Both quantitative metrics and qualitative visualization results demonstrate that WEFNet outperforms existing methods, particularly in extracting complex boundaries and small-scale water bodies.