An Intelligent Approach to Point Cloud Segmentation Based on Cross-Attention in UAV-Based Substation Inspection

In recent years, UAV-based(Unmanned Aerial Vehicle) inspection has become an integral part of power grid monitoring, offering a new avenue for efficient and autonomous assessment of power infrastructure. However, in complex environments such as substations, point cloud data is often sparse and unevenly distributed, posing significant challenges for precise object segmentation. To address these issues, we propose a cross-attention-based Segmentation Network tailored for the segmentation of point clouds obtained by UAVs during substation inspections. The method integrates a multi-scale contextual information fusion mechanism, significantly improving feature extraction for objects of diverse sizes within complex power grid environments. Experiments on real-world power grid datasets demonstrate that our method achieves high segmentation accuracy and performance, making it highly effective for UAV-based power grid inspection tasks. By enabling more precise and efficient drone-based monitoring of power infrastructure, this work advances intelligent power facility surveillance and supports the development of autonomous UAV inspection systems.