With regard to the electric tower identification problem in the high-resolution SAR image under complicated background, it is difficult for the existing target detection algorithm to reach balance in identification precision and efficiency. Therefore, this paper combines the advantages of high YOLOv2 calculation efficiency and high VGG classification precision, and proposes a kind of two-stage target detection algorithm of YOLOv2 and VGG cascade connection by virtue of transfer learning. At the Stage-1, the sliding window and the non-maximum suppression algorithms are combined, and the YOLOv2 is utilized to conduct rapid electric power detection to the whole SAR image, thus obtaining the target detection result of high recall rate; at the Stage-2, the VGG classification model is utilized to conduct secondary classification to the target and background with regard to the target detection result at the Stag-1, thus further eliminating the false positive. This algorithm can be used to enhance the accuracy of only using YOLO v2 f or target identification, thus reducing the false alarm rate of model effectively. The electric towers in the mountainous belt and the plain belt are used as samples for training, thus enhancing the robustness of algorithm. After conducting algorithm testing to the COSMO image in Zhijiang City, Hubei Province, the result shows that the electric tower recall rate can reach 73.8%. The method adopted in this paper can identify the electric tower target of the whole SAR image more accurately through model transfer and two-stage deep learning, thus being further promoted into actual application scene.