Abstract: The conventional aerial target detection method for high-voltage transmission lines is based on the YOLO algorithm to optimize anchor boxes for detection. However, due to its sensitivity to the spatial distribution of aerial targets, the accuracy of bounding box regression is insufficient. Therefore, a lightweight Fast-Unet network is proposed for aerial target detection of high-voltage transmission lines. Using grayscale analysis technology to remove background areas, morphological filtering is applied to determine the region of interest by scanning line by line from the bottom of the image. Input the region of interest into the Fast-Unet network, calculate multiple feature blocks to construct a comprehensive feature vector, and extract target features by combining global and local contrast, color space distribution features. Based on the shape features of the target, a constraint mechanism is constructed, and the target contour is extracted using chain code tracking algorithm. The contour is quantified using Fourier descriptors, and a shape similarity measurement model is established using principal component analysis for detection. The experimental results show that the lightweight Fast-Unet network achieves complete and error free detection of all target areas, with an average accuracy of 95.6%, significantly higher than the 71.2% and 72.8% of the comparison methods, verifying its effectiveness in complex aerial photography scenes.