Change detection (CD) is crucial for accurately obtaining change information from geographic surfaces across different time periods. Recent advances in deep learning have made significant achievements in this field. However, there are still several challenges remaining: (1) Existing methods for change feature fusion are not comprehensive and accurate enough to capture and integrate change information; (2) Traditional CD models often confine themselves to local information and lack the ability to model the global context, which somewhat limits their performance in CD tasks. Therefore, we propose a local-global semantic feature fusion network (LGFNet). This network effectively leverages the long-range modeling capabilities of Transformers and the local perception strengths of CNNs by integrating the global and local features they extract, resulting in efficient feature representation. Additionally, to enhance the change features and reduce redundant pseudo-changes, we designed a feature difference enhancement module. On several benchmark datasets, our method outperforms several state-of-the-art approaches. The experimental results demonstrate that this method detects changes with high accuracy.