Neural Radiance Field Reconstruction Technique Under Layer Training Strategy

The digital twin technique for 3D reconstruction based on virtual models poses challenges for algorithms with high reconstruction rates and high fidelity. In this paper, we propose a NeRF 3D reconstruction method combined with Greedy Layer-Wise Training strategy for the digital twin task of transformers and transmission towers. First, the visible images of power equipment are collected by 360-degree ring photography from UAVs and handheld devices. Secondly, the NeRF reconstruction algorithm is used to form a high-precision and high-fidelity 3D scene model for this dataset; in addition, a greedy hierarchical intelligent training strategy is employed. The experimental results show that the NeRF-based 3D reconstruction model in this paper performs very well in the reconstruction tasks of power transformers and transmission towers, and can highly restore the 3D morphology and texture of the objects, while having strong stability and robustness.