Neural radiance field (NeRF) has emerged as a cutting-edge approach in neural rendering for 3D scene representation. Addressing the limitations of conventional NeRF in handling dynamic foregrounds and depth discrimination in complex power grid scenes, in this study, we introduce an enhanced NeRF algorithm leveraging the dense prediction transformer (DPT). Our method employs DPT to eliminate dynamic foreground elements across viewpoints and integrates a dual-sphere reparameterization with a differentiable sampling strategy within the NeRF model to apply blur constraints to distant scenes. This approach notably improves the reconstruction quality of power grid towers. Experiments show that our algorithm surpasses traditional NeRF variants in capturing details, rendering backgrounds, and overall visual quality, with improved performance in peak signal-to-noise ratio mean structural similarity index measure, and multi-scale structural similarity index measure (MS-SSIM) metrics compared with state-of-the-art models.