Low-altitude Obstacle Model Identification Method for Power Transmission Corridor Based on Improved 3D Point Cloud Feature Fusion

Because of the physical characteristics of high voltage, the transmission system needs to occupy a long air corridor, forming a power transmission line without a closed environment. The space not only covers transmission equipment and lines, but also may be invaded by other obstacles. For example, with the growth of trees, branches often invade transmission lines, causing short circuits and other power accidents. Therefore, for such elements as houses and buildings, sufficient safety distance should be reserved in the design and construction stage to reduce the possibility of power accidents. The lidar sensor carried by UAV can collect point cloud data, which provides coordinate information related to the power background (such as power lines, and towers) in remote sensing images. In this paper, the point cloud data is preprocessed to effectively reduce the amount of data and standardize the coupling of heterogeneous data to simplify the complexity of modeling. By constructing the probability model of point cloud feature differences and neighborhood spatial features, various obstacle features of the transmission corridor are successfully identified. The DGCNN (Dynamic Graph CNN) simulation algorithm and the region growing strategy are used to extract the features of the target model, and the adaptive density clustering is used to distinguish the factors of geometric and morphological features, thereby identifying relative differences. Finally, the point cloud data is used for accurate classification. The accuracy index of three-dimensional point cloud data is $\mathbf{1 6 . 7 \%}$ higher than that of the comparison method. It is judged as the best choice in the definition-branch recognition module. Under the condition of meeting the growth conditions, this improvement will minimize the identification error of three-dimensional point cloud images, and has the potential to migrate from two-dimensional image processing to three-dimensional point cloud processing, which provides technical support for solving the problem of low-altitude obstacles in transmission corridors.