Research Of Running State Detection System For Disconnecting Switch Based On Machine Vision

With the rapid development of substation inspection robot and artificial intelligence technology,Machine vision plays a more and more important role in the inspection process of electrical equipment such as disconnecting switch.Conventional detection methods based on machine vision use image segmentation and line extraction to detect the running state of disconnecting switch.Although such methods are simple to realize,the demand for acquisition angle and distance is very high while collecting images of disconnecting switch,and it is difficult to locate the switch region accurately.At the same time,there is a shortcoming that the segmentation threshold is difficult to select automatically and the recognition accuracy is not high.In view of the above problems,the method and detecting system of running state detection for disconnecting switch based on feature matching and random forest classification are studied.An improved SIFT(Scale Invariant Feature Transform)algorithm is applied to complete the coarse recognition of running state of single-phase disconnecting switch.The classifier model build by random forest is used to accurately recognize the running state of three-phase disconnecting switch.The main contents and conclusions are as follows:According to the trait that point features of images are not susceptible to the impact of factors such as rotation and scale,a coarse recognition algorithm of running state based on improved SIFT is designed.The dimensionality of feature descriptor is reduced by improved neighborhood division,and improved matching strategy is adopted to eliminate error matching points.The experimental results show that the recognition rate of improved algorithm is over 95% with matching time reducing by about 50%.Compared with primal algorithm,the speed and accuracy of matching recognition can be both improved effectively.In view of the fact that coarse recognition algorithm cannot detect the running state of multi-disconnecting switch,an accurate recognition algorithm of running state based on random forest classification is designed.The validity of the feature description is enhanced by using histogram of oriented gradient and edge features as the feature set of disconnecting switch.The switch target is accurately positioned by combining generalized Hough transform with traditional decision tree algorithm.Particle Swarm algorithm is used to optimize the training parameters while training switch model by random forest,which improves the training effect of switch model and ensures the best detection efficiency.The detection platform based on TMS320DM642 and man-machine interface based on MFC(Microsoft Foundation Class)are designed to integrate the detection system at last.The test results show that the total recognition rate of running state of three-phase disconnecting switch is over 96% under various disturbances and changes in the angle of view,and the time-consuming of detection takes only about 3.2s.