Abstract: In order to achieve the automatic docking assembly of the transmission tower flanges, a positioning method combining deep learning and traditional image processing was proposed. Initially, the positions of the bolt holes in the images were delineated through an improved Faster-RCNN model. Subsequently, the image of the delineated area was processed using grayscale conversion, filtering, Canny edge detection, and edge selection. The two-dimensional coordinates of the center of the flange bolt holes were obtained by fitting using the method of least squares. Lastly, the three-dimensional coordinates of the flange bolt hole centers were acquired by combining a binocular depth camera. Experimental results demonstrate that the detection accuracy of the improved model reaches 95% under the condition of IoU=0.75, with an average detection time of 0.147 seconds. The positioning error is controlled within 0.91 millimeters. It is concluded that the proposed positioning method can precisely acquire the three-dimensional coordinates of the center of the tower tube flange bolt holes, providing accurate coordinate information for the automatic docking assembly process of transmission towers.