Accurate recognition and segmentation of insulator image is an important prerequisite for state perception and defect diagnosis of overhead power line insulators. In this paper, a two-stage image recognition and segmentation method combining the Yolo (you only look once) v5 and Grabcut is proposed for the visible image of insulators. This paper firstly collects images, establishes data set and trains Yolo v5 to realize insulator recognition. Then the recognition frame coordinates are used to determine the region of interest, predict the foreground and background, and realize the adaptive segmentation of insulators based on the Grabcut method. This method is used to identify and segment the on-line monitoring visible light image of overhead power line insulators. The results show that this method can accurately locate the edge of insulator under complex background and segment insulator without segmentation and annotation or manual interaction, which can efficiently improve the efficiency of insulator image analysis.