In response to the issues of excessive manual intervention and low efficiency in the unmanned aerial vehicle (UAV) inspection of overhead lines of distribution network, this paper proposes a method for autonomous inspection and attitude control of UAVs based on multi-sensor fusion. The method employs a deep learning model based on MinkUNet for segmentation of corridor point cloud data followed by modelling of power poles and transmission lines and three-dimensional flight path planning to achieve the automatic planning of UAV inspection path. During the flight execution phase, filtered millimeter wave radar data and depth camera data are fused under a Bayesian theory-based occupancy grid map update strategy to detect newly emerged obstacles. An improved artificial potential field method is utilized for autonomous obstacle avoidance. To compensate the locating drifts when positioning signals are disrupted, a lightweight YOLOv8-based target recognition model and gimbals control strategy are introduced to enable precision alignment and photography of the inspection targets. The experimental results based on real-world and simulated environments demonstrate that the proposed methods are effective and robust for complex power distribution network inspection scenarios.