| Pantograph is a key component installed on the top of high-speed railway(HSR)to obtain electrical energy from the catenary.The electrical energy required by HSR is obtained by friction between the pantograph and the catenary.Pantograph is one of the key components to ensure the safe operation of HSR,and once the pantograph fails,it will directly affect the operation safety of HSR.Based on the existing pantograph video monitoring system,it is very important to study the real-time online machine vision algorithm to realize the real-time detection of pantograph fault and status,and then accurately evaluate the current health status of pantograph to ensure the safe and stable operation of HSR.For the current methods of real time pantograph detection through machine vision,the traditional image detection method is only applicable to pantograph structure detection with simple background and single scene,while it is difficult to establish a sample set covering all complex scenes and actual operation conditions when using neural network for pantograph structure detection due to the small number of pantograph structure abnormal samples,many actual operation scenes,and interference from outside full of uncertainty.It is difficult to establish a sample set that covers all complex scenarios and actual operation conditions,which leads to more misjudgments when structural anomalies occur or external disturbances are encountered.Therefore,in order to solve the following problems,the main research contents of this paper are as follows.(1)To address the problem that the current pantograph structure anomaly samples are small and the sample set for training neural network is not rich enough,this paper proposes to use DCGAN to expand the kinds of data samples,and then use YOLO object detection algorithm to realize real-time localization of pantograph area.Compared with a series of requirements(e.g.,clear pictures,simple background,less interference,etc.)when implementing pantograph detection based on traditional image methods,the use of YOLO achieves higher accuracy and better environmental adaptability for real-time accurate localization of pantograph regions.At the same time,DCGAN enriches the variety of samples and further improves the accuracy of the trained model.(2)To address the limitations and low accuracy of the current pantograph structure abnormality detection method,this paper fits the pantograph characteristic curve by polynomial curve fitting based on the pantograph area located,and then compares it with the curve obtained by normal pantograph fitting through the improved LCSS algorithm to achieve real-time accurate judgment of the pantograph structure.The improved LCSS algorithm can not only accurately detect and measure abnormalities when the pantograph structure is abnormal,but also accurately locate the area where the abnormality occurs,while still performing well under the influence of complex scenes and external environmental interference.(3)To address the problems that the current pantograph detection methods are easily influenced by the environment,cannot effectively cope with the interference of external scenes,and have low accuracy and cannot meet the actual operation requirements of HSR,this paper fully considers the impact of the complex background interference and fuzzy dirty lens that may be faced during the operation of HSR,and designs a pantograph immunity model for HSR that combines the complex background exclusion algorithm and the fuzzy dirty detection algorithm to improve the accuracy of pantograph detection under the relevant scenes and reduce a series of negative impacts brought by these external interferences on pantograph detection.The paper concludes with a summary and outlook.The algorithm proposed in this paper has been fully tested and validated on different datasets with good performance,and the results show that the algorithm meets the actual operational requirements of HSR and has high practical application value. |
PDF Full Text Request