The main form of power transmission in China is the high voltage transmission line transmission.In order to ensure the safety and stability of transmission lines,it is necessary to carry out regular inspection of transmission lines.The current line inspection mode has got rid of manual inspection,and UAV inspection or robot inspection is mainly used.In order to solve the problem of automatic patrolling on transmission wire,the patrolling robot must be able to recognize the transmission line fittings on the wire.The robot can realize automated reciprocating inspection by passes over the recognized fittings with the preset action of obstacle crossing.Based on the identification requirements of the patrolling robot for various kinds of power fittings on the line,this paper proposes a scheme for the patrolling robot to recognize the power fittings on the 110kV overhead high-voltage transmission line based on the analysis of the structure and image characteristics of the common power fittings.And according to the application environment of the line inspection robot,optimized the selective search method on ROI generation.The main work contents of this paper are as follows:(1)Based on the pictures collected from the Internet and the photos of the purchased power fittings,a sample set was made for the training of the power fittings recognition model.This paper analyzes the shape and structure characteristics of several common power fittings,including vibration damper,insulator chain and strain clamp.(2)Considering the structural characteristics of several kinds of fittings and the actual engineering application environment,analyzed the design requirements and carried out the model scheme.The scheme includes the extraction of HOG feature to establish the sample data set,and the design of DNN network and CNN network structure.(3)According to the designed recognition model,the collected sample data set is used for training and testing,and it is verified that DNN+HOG model is more suitable for the recognition of power fittings in this environment compared with the CNN network model.Then verifyed the feasibility and rationality of the recognition technology. |