Abstract: As a crucial component of electricity resource allocation, the distribution network plays a significant role at the end of the information power system by ensuring the safe transmission of electrical energy to end users and proper power distribution. Due to the characteristics of main grid power transmission, distribution network lines are often concentrated in remote suburban areas. The conventional method of manual inspection is time-consuming, resource-intensive, and inefficient, resulting in poor timeliness. Therefore, the current research focus in intelligent distribution network construction is on unmanned aerial vehicle (UAV) automatic inspection methods based on 3D reconstruction. However, in the context of distribution network channels, the application of 3D reconstruction has mainly focused on pose estimation and reconstruction algorithms, neglecting the influence of feature matching on pose estimation and map reconstruction. This paper proposes the construction of a dataset for the distribution network channel environment and achieves dynamic feature matching tasks in complex environments by training the feature matcher model MatchFormer, providing technical support for 3D reconstruction in distribution network channels.