Research Of Catenary Image Detection Based On Machine Learning

In recent years,the rapid development of high-speed railway has brought heavier catenary detection works.Traditional manual detection fails to meet the demand of daily inspection from infrastructure manager,the development of real-time online automatic detection technology is very urgent.At present,although there are quite many researches and papers about catenary image detection,practical applications are still rare,or the effect is not satisfactory.The complexity of the catenary device and environment changes,make the image processing and recognition even difficult,especially in high-speed conditions.In this paper,the algorithm of machine learning as well as support vector machine(SVM)classifier is used to complete the application research of catenary image detection.The principle of SVM and the theory of image feature extraction are introduced.The detections of pantograph,insulator on catenary and the measure of catenary localizer slope are taken as examples to illustrate the concrete application of the SVM by description of program processes.Finally,C++ and OPENCV library are used to verify the experiment.The experimental results are as follows: During the image detection of carbon skateboard under catenary,SVM can achieve high recall rate with sacrifice of the detection accuracy,which can meet the safety requirement of the detection.In the case of insulator detection,the classification effects of different SVMs are compared using different kernel functions.The Gaussian kernel function is suitable for the vector machine based on the gray level co-occurrence matrix;And in the measure of catenary localizer slope,SVM can instantly distinguish right locations of catenary's localizer,with which the program can run faster.It can be seen that SVM has high application value in catenary image detection.The key lies in: 1.Proper solution of the image segmentation of components;2.Selection of the most suitable image feature for extraction;3.Plenty data to supportthe sample library.With these problems solved,SVM can quickly identify the fault picture,so that to achieve real-time automatic detection.