Transmission line is an important power equipment, but long time exposed to the natural environment operation will inevitably produce various defects, if these defects are not found and repaired in time, it will bring great hidden danger to the stable operation of the power system. Therefore, it is very important to conduct regular inspections of transmission lines and troubleshoot defects. It is time consuming, laborious and subjective to recognize the transmission line image manually, so we need to use computer technology to assist the inspection of transmission line defects. In this paper, we proposed a one-stage object detection network AG-RetinaNet which is used to identify broken strands and foreign bodies in transmission line. Experiments demonstrate that in the multi-objects detection of broken strands and foreign bodies, it achieves the best detection effect compared with Faster R-CNN, YOLOv3, YOLO v5 and RetinaNet.