Abstract:At present, the degree of informatization of substation equipment is constantly improving, and higher digital requirements are put forward for each module of secondary equipment.In order to solve the problems of complicated, time-consuming and error-prone manual identification procedures for the state of the secondary equipment, improve the inspection efficiency of the pressure plate, and maintain the safe and stable operation of the secondary equipment,this paper studies and compares the difficulty and accuracy of machine learning algorithms and deep learning target detection algorithms,and finally chooses the YOLOv3 algorithms.After converting the platen standard state document into json data, use the imagenet-based YOLOv3-darknet53 pre-training model and import the platen switch image sample set.After labeling the sample set with LabelImg, perform deep learning to obtain the trained model. Use this model to obtain the identification data of the on-site platen diagram, compare the standard state data with the actual picture data and draw the graphics, so as to realize the accurate identification and labeling of the platen switch state.The results show that the success rate of pressure plate switch position and switch status recognition reaches 99%, which effectively improves the work efficiency and accuracy of pressure plate inspection, and greatly improves the digitization level of secondary equipment.