Abstract: Objective: To automatically identify the fracture state of rotary double ear pieces for rail transit catenary support device, it is necessary to develop adequate detection methods for identification, and enable timely replacement and maintenance in case of equipment failures, thereby improving the stability of catenary system arm structure. Method: Firstly, the global image of catenary support device is preclassified using Hough transform, dividing the basic elements of the image for detection into three categories: rodshaped objects, rotary double ears and insulators. Then, the category of rotary double ears are primarily identified using affine invariant moments; subsequently, the rotary double ears are accurately located; and finally, the fracture features of the ears in local images are identified through the distribution of extremal parameter matrices. Result & Conclusion: The state detection method based on machine vision is proposed, which can accurately identify the fault features of support device rotary doubleear pieces, with shorter detection time and higher identification rate. Specifically, for the key detection sections, offline reevaluation can be conducted based on this method through a data processing center to meet the system reliability requirements. This method can fulfill the detection and monitoring needs of catenary components regardless of online or offline detection.