Fault Detection And Identification Of Cable Leakage Fixture In Tunnel Based On Deep Learning

The railway tunnel communication network coverage mainly depends on the leaky coaxial cable,fixed on the tunnel wall by the fixtures.The fixtures are easy to loosen or even fall off due to the impact of air pressure and energy wave produced by highspeed trains for a long time,which seriously affects railway operation safety.At present,fixture detection mainly relies on video monitoring.Surveillance cameras are collected uninterrupted along the train for a long time,generating many videos along the railway.How to detect and identify fault fixtures quickly and accurately has an urgent practical demand.Under the support of the Natural Science Foundation of Hebei Province and the enterprise entrusted project,this paper carried out the research on fault detection and identification of tunnel leaky cable fixtures based on deep learning.The innovative research results obtained and the main work completed are as follows:(1)A tunnel boundary extraction algorithm based on spatio-temporal diagram is proposed.Because the tunnel is only a tiny part of the video along the railway,this paper proposes a tunnel boundary extraction algorithm based on the spatio-temporal diagram.This algorithm can improve detection efficiency and can be used to separate the tunnel parts.Firstly,video frames along the railway are fused into a sequence of pixels,and the sequence of pixels is spliced into a spatio-temporal diagram along the time axis.Then the spatio-temporal diagram is binarized and the convolution filter is constructed to denoise.Finally,the boundary frame number of the tunnel is extracted by vertical projection,and the tunnel sequence is corrected by the double marker bit method,which realizes the precise extraction of the tunnel boundary.Experimental results show that the recall rate and precision rate of the proposed algorithm are 99.14 % and 98.87 %,which are obviously better than the contrast method.In addition,the proposed algorithm has very low computational complexity because it only deals with one list of pixels of the video frame.(2)A fault detection and identification algorithm for tunnel leaky cable fixtures based on an improved SSD algorithm is proposed.Aiming at the slow detection speed and low recognition accuracy of tunnel leaky cable fixture,this paper proposes a fault detection and recognition algorithm of tunnel leaky cable fixture based on an improved SSD algorithm.This algorithm increases the network width by combining the Inception structure.The residual structure is used to improve the depth of the network and optimize the network depth structure.A large number of deep separable convolution and 1×1 convolution structures are used to reduce the number of model parameters and improve model detection and recognition efficiency.Experimental results show that the detection speed of the proposed algorithm is 26.6 fps,and the average accuracy of fixture identification is 86.6 %,which is obviously better than the original SSD algorithm and the Mobile Net SSD algorithm.(3)A fault detection and recognition algorithm for tunnel leaky cable fixtures based on feature enhanced SSD is proposed.Accurate detection and identification of fixture faults are significant for the safe operation of railways.Therefore,this paper proposes a fault detection and identification algorithm of tunnel leaky cable fixtures based on feature enhanced SSD to improve the accuracy.Firstly,the width and depth of the feature extraction network are increased to enhance feature extraction capability.Then,the feature fusion method was adopted to make the layer to be detected fuse the feature information of the high level network and the low level network,to enhance the feature information of the layer to be detected and improve the detection accuracy of the model.Finally,an attention mechanism is introduced to enhance the importance of key features and ignore redundant feature information,to improve network performance.The experimental results show that the average accuracy of this algorithm is 90.4 %,which is obviously better than the contrast algorithm.(4)A prototype system for fault detection and identification of cable leakage fixtures in the tunnel is built.This paper builds a tunnel cable leakage fixture fault detection and recognition prototype system based on feature enhanced SSD and Tensor Flow deep learning framework for the application scenario of tunnel cable leakage fixture fault detection and recognition.The system includes two main functions: image detection and video detection,which can significantly improve fixture detection and recognition efficiency.The system test shows that the system basically meets the actual needs of fixture detection.