Research on Automatic Extraction Technology of Power Transmission Tower Based on SAR Image and Deep Learning Technology

In recent years, extreme weather events such as heavy snowfall and floods have occurred frequently, posing great challenges to the safe operation of transmission lines. In order to quickly assess the disaster situation of the transmission tower, it is necessary to use the synthetic aperture radar (SAR) technology to quick survey the large-scale transmission line conditions, and to locate and identify hidden dangers of large-scale transmission towers. At present, for the identification of electric towers in high-resolution SAR images in complex background, the existing target detection algorithms have shortcomings in the accuracy and efficiency of electric tower identification. To this end, this paper proposes a two-stage target detection algorithm that combines YOLO v2 and VGG model cascades with migration learning. In the Stage-1 phase, combined sliding window and non-maximum suppression algorithm, YOLO v2 is used to perform fast electric tower detection on the whole scene SAR image, and the target detection result with high recall rate is obtained. In the Stage-2 stage, the VGG classification model is used to perform secondary classification of the target and background of the target detection result in the Stage-1 stage, further eliminating false positives. The algorithm can improve the accuracy of target recognition using YOLO v2 alone, and effectively reduce the false alarm rate of the model.