Abstract: To effectively reduce the omission rate of automated statistics of distribution network line assets, this paper proposes an automated statistical technique for power assets of distribution network power supply lines. This technology calculates the edge gradient amplitude of distribution network asset images such as wires and towers, and uses Sobel algorithm to separate the foreground of asset images; using ICP algorithm to achieve precise registration of digital images and 3D point cloud data; using a feature pyramid network, extract and fuse the deep and shallow features of distribution network assets; combining CNN model training to predict the asset type to which the input image belongs and obtain asset type labels; finally, count the number of assets under each category and output an automated statistical list. The test results show that when using this method to statistically analyze regional power assets, the dynamic deviation rate of asset omission is relatively low, significantly reducing the degree of asset omission.