Abstract
In order to achieve rapid and accurate estimation of power project quantities, this paper proposes a 3D automated calculation method that integrates drone photogrammetry and deep learning. By introducing a catenary line constraint to improve the reconstruction integrity of small targets in the SfM algorithm, and using a RandLA-Net network with a channel attention mechanism and Focal Loss function to optimize point cloud segmentation accuracy, a parametric rule library is constructed for automatic quantity extraction. Experimental results show that for the estimation of line length, earthwork volume, and tower material weight, the relative errors of this method are 1.2%, 3.2%, and 2.5%, respectively, showing significant improvement in accuracy compared to 2D digital methods. The processing efficiency is 2.5 hours per 15 kilometers, which is nearly six times faster than manual measurement. When the point cloud density exceeds 200 points/m², the tower recognition accuracy reaches 96.5%. This method provides a reliable automatic quantity estimation solution for power engineering, with excellent results in flat areas, though further optimization is needed in complex terrain conditions.