Foreign Object Detection Of Railway Catenary Based On UAV

Catenary system plays an important role in the operation of electrified railway.At present,the inspection of railway catenary mainly depends on manual inspection,which has low efficiency,high labor cost,and is affected by personal subjectivity,it may cause problems such as false detection and omission.With the application and development of image processing and machine vision related technologies in the field of UAV.Using UAV to inspect the traction power supply system can effectively ensure the safe operation of the traction power supply system,reduce the inspection cost and improve the inspection efficiency,which is of great significance to promote the intellectualization of inspection.The thesis studies the detection of foreign objects in catenary based on UAV,and uses DJI UAV to capture catenary images to realize the automatic detection of foreign objects in catenary.The main research work of the thesis is as follows:1.According to the experimental requirements,the thesis made a comparative analysis of DJI Mavic 2 Professional version and DJI Air 2S,and finally selected DJI Mavic 2 Professional version for data collection.The UAV flew at an altitude of 4200mm～4800mm from the rail surface,and collected 50 real videos of no-foreign objects on the catenary and foreign objects on the catenary respectively,from which 6120 images are selected.The experimental data set and test set are divided in a ratio of 9:1.Label Img software is used to mark catenary poles and foreign objects.2.In order to meet the use requirements of portable devices,an improved SSD algorithm based on Mobile Net V2+FPN+Attention module is proposed.The VGG-16 network with many parameters and large computation is replaced with Mobile Net V2,and FPN is introduced to solve the problem of poor detection effect of small objects.In order to further improve the discriminability of the features extracted from the network,a hybrid attention module is introduced in the network,which enables the network to pay more attention to useful information.Meanwhile,the influence of redundant information on feature extraction is suppressed and the detection accuracy is further improved.3.In the thesis,the DJI Mavic 2 Pro is used to capture catenary images in the training field of Jilin Railway Technology College to realize automatic detection of foreign objects in the catenary.By comparing and analyzing Fast R-CNN algorithm and SSD algorithm from the dimensions of m AP,macro F1-score,FPS and parameter quantity,the m AP value of the algorithm in the thesis reaches 62.63%,which is 2.48% higher than the Fast R-CNN algorithm and 0.83% higher than the SSD algorithm.The macro F1-score value of this algorithm is 76.47%,which is 2.41% higher than the Fast R-CNN algorithm and 0.89% higher than the SSD algorithm.Meanwhile,the speed of the algorithm in the thesis reaches 124.71 FPS,which is 11 times faster than the Fast R-CNN algorithm and nearly 1 times higher than the SSD algorithm.It can be seen that the thesis reduces the amount of parameters while improving the detection accuracy,thereby improving the speed of calculation.