Identifying power transmission towers (PTTs) in synthetic aperture radar (SAR) images is challenging due to the unique imaging mechanism. In high-resolution SAR images, background clutter and intricate geometric features of the towers compromise the accuracy of current PTT detection techniques. This paper proposes a new method for detecting PTTs in high-resolution SAR images using a single-stage architecture. First, rotated bounding boxes are used to avoid background noise. Second, the Pyramid Vision Transformer (PVT) encoder is employed to learn various geometric information about the PTTs. Finally, we introduce the adaptive training sample selection (ATSS) and statistical features to distinguish positive and negative samples automatically. The effectiveness of the proposed method is confirmed by the experiments based on high-resolution (better than 1m) SAR images from Capella Space open data.