Overhead transmission line inspection is an important task in power grid maintenance, and the utilization of unmanned aerial vehicles (UAVs) for line inspection has become a significant approach in power inspection operations. Firstly, the overview of the architecture of the human-machine collaborative operation system and the UAV intelligent autonomous operation system in UAV inspection tasks are provided. Next, the current status of datasets for defect inspection in overhead transmission lines is analyzed and the data augmentation techniques are discussed. Subsequently, this paper reviews typical deep learning-based methods for UAV image defect detection in detail, along with evaluation metrics. The advantages and limitations of various approaches are compared and summarized. Furthermore, the impact of image acquisition specifications, dataset formats, and specialized defect detection algorithms are discussed on the detection performance for overhead line defects in UAV image visual inspection methods. The shortcomings of image detection metrics and category definitions in the specialized field of power inspection are pointed out. Finally, future directions for deep learning-based UAV image defect detection tasks are explored.