For key low texture components of the transmission tower, most detection methods rely on the labeling and training of a largenumber of sample images, while these samples are usually scarce. Proposes a transferable deteetion method-PowerNet for the key lowtextured components. It combines Siamese network and eross-correlation convolution to fuse the features of sample block and searchregion,and effectively filters out the background of the component sample image using mask eropping, Finally, scoring candidate regionis proposed where the seale, location, and loU are adopted to improve the accuracy of detection. Experiments show that mask ermppingand sooring candidate region play an important role in accurate detection. According to AP⁵0 eriterion, the avernge accuracy rate of ourdeteetion method reaches 98%, which can detect various eomponents of the transmission tower concerning the varianee of view angle,scale, or illumination.