Abstract: Water body identification is of significant value for water resource management and aquatic environment monitoring, yet current models often suffer from low accuracy and poor performance in water edge extraction. To address these challenges, this study proposes a novel water body identification model, MTMFunet. The model first performs a coarse extraction of water bodies from remote sensing images using the U-Net convolutional neural network. Subsequently, it utilizes the Mixture Tuned Matched Filtering(MTMF) model for mixed pixel decomposition to precisely extract water from mixed pixels along water edges. This multi-step process involves acquiring endmember spectra, extracting mixed pixels, determining relative endmember abundances, and inverting water sub-pixels, thereby enhancing the model's overall accuracy. Finally, a comparative experiment was conducted between the MTMFunet model and the standard U-Net model. The MTMFunet model achieved an accuracy, precision, recall, F1-score, and Kappa coefficient of 99.98%, 99.94%, 99.90%, 99.92%, and 97.60%, respectively. Notably, its Kappa coefficient, a key indicator of model accuracy, was 7.83% higher than that of the U-Net model. These results demonstrate that the MTMFunet model exhibits superior capabilities in water edge extraction and overall water body identification accuracy.