electric railway polymeric insulator

    Electrified railway insulators are important equipment in the catenary support suspension device, and their status affects the stability of the traction power supply system. Therefore, it is necessary to study real-time and intelligent status detection methods. Accurate and efficient identification of insulators is the premise to realize intelligent detection of insulator bad state, ensuring that subsequent detection work is carried out in the target local image, reducing the difficulty of state detection and the calculation of detection algorithms, thereby improving the speed of detection and the accuracy of detection results. This paper takes the high-efficiency and clear images collected by the railway comprehensive inspection vehicle as a sample, and applies the related technologies of computer vision to realize the efficient identification of electrified railway insulators. First, the main two types of target recognition methods are introduced: template matching and statistical pattern recognition; then feature-based image matching method is used to realize insulator recognition; finally, statistical pattern recognition method is used to extract three characteristics of insulators: Haar, LBP and HOG. After training, the obtained classifier is used for insulator identification. For the identification of electrified railway insulators by template matching, this paper uses three feature matching methods based on SIFT, SURF and ORB to realize insulator identification, and analyzes and compares the identification effects of the three methods, and concludes that the comprehensive performance of SURF is better. . In order to compare the recognition effects of the three matching methods, a friendly MFC interface that can fully realize the recognition process is designed. For the statistical pattern recognition method, the AdaBoost algorithm and SVM are used to train the classifier. For Haar features and LBP features, the AdaBoost algorithm is used for classifier training, and through the evaluation of the generated classifiers, it is concluded that changing the interception method, size and number of positive and negative samples can significantly improve the performance of the classifier. For the HOG feature, SVM is used to train the classifier, and changing the interception method, size and number of positive and negative samples can improve the recognition effect to a certain extent. In the experiment, in the environment of Visual Studio 2013 and OpenCV3.0, the identification of electrified railway insulators is realized by programming. The experimental results show that for the insulator recognition based on image matching, the recognition performance can be significantly improved by eliminating the mismatch. For the insulator identification based on the statistical pattern recognition method, changing the interception method, size and number of positive and negative samples can significantly improve the identification effect. With appropriate sample interception method, size and quantity, a LBP feature cascade classifier is obtained, which has high accuracy and speed.