In view of the practical problems of small proportion and close interval and difficult to identify accurately of ground wire area in the images obtained by handheld devices when the temporary ground wire is attached to the transmission line, this paper proposes an improved Faster R-CNN method to realize ground wire target recognition. the low- and high-level feature maps of the convolutional network are fed into the RPN on the basis of the original Faster R-CNN method to achieve multi-scale target detection, and the non-maximum suppression is improved. The improved model is transplanted to the handheld data acquisition device. Through simulation and field test, the detection accuracy of grounding wire is 94.8%, which is 7.5% higher than the original method. It shows that the proposed method can effectively improve the overall performance of target recognition.