To address the problems of traditional methods in 3D reconstruction of ancient architecture

including low efficiency

difficulty in processing complex textures

and geometric distortion caused by repetitive textures

an improved method based on depth-supervised neural radiance field(NeRF)is proposed using images of ancient architecture captured by unmanned aerial vehicle(UAV) aerial photography sensors.Truncated signed distance function (TSDF)is introduced to enhance geometric representation

and depth data obtained from sparse point clouds is used to supervise neural network training

thereby improving geometric reconstruction accuracy.A progressive training strategy is used to optimize hash table representation and mitigate hash collisions.Experimental results demonstrate that the reconstructed images of the proposed method possess clearer structural details than current mainstream methods

and the peak signal-to-noise ratio(PSNR) and structural similarity index(SSIM) are increased by 15% and 10% on average

respectively.This method achieves rapid 3D reconstruction of ancient architecture while significantly improving the geometric accuracy of reconstruction

and can generate higher-quality scene surface mapping models.