Automatic detection method of abnormal targets in railway catenary based on multi-scale coupled convolutional network

Abstract

The invention discloses an automatic detection method for abnormal targets of railway catenary network based on multi-scale coupling convolution network, and obtains multiple final candidate areas of the image; In the convolutional network, parameter training and feature extraction are performed; the extracted features are input into the extreme learning machine ELM classifier to classify the extracted features; according to the position of the candidate frame obtained, the position of the candidate frame is corrected by using a regressor, so that The position of the abnormal target in the image information is obtained, that is, the position of the corrected candidate frame. The invention is suitable for automatic detection of abnormal targets in railway catenary, can obtain more accurate abnormal detection effect than human eye observation, has high detection precision and high degree of automation.

Description

Technical field [0001] The present invention relates to the field of railway safety protection, and more particularly to an automatic detection method of an abnormal target based on a multi-scale coupled convolutionary web. Background technique [0002] The electrified railway is generally powered by a high-frame cable. The safety issues of contact network directly affect the operation of the railway train, and the abnormalities such as the Bird's Nest are an important risk source that directly threatens the safe and reliable operation of the railway power line. Currently, it is necessary to discover and remove an abnormal goal by manual inspection, not only is wasteful, not only to troubleshoot safety hazards in time. In order to overcome the above problems, the abnormal target detection of the railway contact network is carried out. However, research on automatic detection of abnormal targets has just started, and the accuracy and efficiency of abnormal detection cannot meet th...

AI Technical Summary

Problems solved by technology

However, for a computer, facing an RGB pixel matrix, it is difficult to directly obtain an abstract concept such as a dog or a cat from the image and locate its position, and sometimes multiple objects and messy backgrounds are mixed together, Object detection is more difficult

[0004] For the research on the automatic detection of abnormal objects, there are three main problems in the detection of abnormal objects in the traditional method. One is that the area selection strategy of the sliding window is not targeted; the other is that the manually designed features are not very robust to the change of diversity. Third, the accuracy of automatic detection of abnormal targets needs to be further optimized, and the efficiency is far from meeting the requirements of real-time applications

Images