Region-aware neural radiance fields for three-dimensional building reconstruction and facade detail preservation

High-fidelity textured three-dimensional (3D) building meshes are essential for digital twins, facade inspection, and condition assessment, but reconstructing fine details such as cracks remains challenging. This paper proposes a region-aware Neural Radiance Field (NeRF) method for reconstructing building meshes. Key innovations include: (1) A region-aware adaptive sampling algorithm that increases sampling density in complex texture areas using a 3D spatial attention field; (2) A decoupled pipeline separating mesh extraction from texture synthesis via learned signed distance fields; and (3) region-guided mesh refinement that enhances geometric detail in critical facade areas. Experiments on UAV-captured building data show better mesh regularity and texture clarity than representative photogrammetric and neural baselines. ROI-based metrics and local geometric-detail indicators provide complementary evidence of improved reconstruction in fine-textured regions. The workflow yields an explicit textured mesh without dense point-cloud meshing after camera pose estimation, supporting offline digital documentation for inspection-oriented building models.