Image inpainting is a challenging computer vision task that focuses on reconstructing missing or damaged regions in images. Existing image inpainting methods often produce overly smooth or blurred results for medium-to-large occlusions due to limited structural reconstruction, even when guided by sketch priors. To address these issues, this paper proposes Sketch-guided Progressive image Inpainting (SketchPI), which progressively restores missing regions with refined structural guidance. First, we generate more accurate sketches by designing a Dynamic Adaptive Selective Edge Deformation (DASED) module, providing richer details for the filling regions. Second, we introduce a Multi-dimensional Semantic-aware Feature Fusion Module (MSFFM) for image reconstruction, designed to address feature misalignment caused by irregular mask boundaries and feature redundancy or distortion. Combined with the GAN framework, our model produces structural soundness and photorealistic detail fidelity. Extensive experiments demonstrate that our method achieves robust edge alignment and visually plausible textures under larger and irregular masks than existing state-of-the-art image inpainting methods.