A method based on unsupervised domain adaptation was proposed to enhance the accuracy of power line detection by drones in low-altitude environments. The method effectively reduced the domain gap between the source （simulated data） and target （real-world data） domains by transforming style features in the shallow layers of the network. Specifically， a style-adaptive instance normalization technique was designed， which could adjust the style characteristics of source domain data while preserving semantic information， making it more similar to target domain data. The experimental results showed that the method proposed significantly improved performance in power line semantic segmentation tasks， achieving an approximate 4% increase in Intersection over Union （IoU） compared to the previous state-of-the-art methods. It provided effective technical support for unmanned aerial vehicle intelligent obstacle avoidance， reducing flight risks associated with power lines and lowering the cost of data annotation.