Research On Visual Recognition Of Casting Number Of Railway Freight Car Couplers

Because of its simple and exquisite mechanical structure and reliable safety performance,the train coupler is widely used for the connection between carriages.As an important component of the train,the efficiency of its maintenance method and the accuracy of the maintenance results are particularly important.At present,the automation of the coupler production and maintenance site is relatively low.In order to make the inspection result correspond to the coupler identity one by one,workers need to manually record the coupler number on the paper inspection card or input it into the coupler inspection system.Such operation mode is time-consuming and laborious due to the dark light on the site,low number contrast,various arrangement forms and other reasons,and is prone to errors in number reading and recording due to visual fatigue of workers,and the accuracy is difficult to guarantee.With machine vision technology and optical character recognition(OCR)technology becoming more and more popular in actual production and life.In this paper,the method based on machine vision is used to replace the traditional manual method to recognize the coupler number.With the goal of improving the recognition accuracy and efficiency of the coupler number,the detection and recognition algorithm of the coupler number is studied to further promote the automation and intelligence of the coupler inspection and repair production line.The main research contents of this paper are as follows:(1)Aiming at the problem that the detection accuracy of coupler circular arrangement number is lower than that of linear arrangement number,by analyzing the distribution law of circular arrangement number and the surrounding geometric shape,a character correction method of improved Hough transform circle detection is proposed,which transforms the coupler circular arrangement number into linear arrangement number,and realizes the preliminary improvement of the recognition accuracy of circular arrangement number and the unification of number arrangement mode.(2)Aiming at the problem that the CTPN model is not accurate enough to detect the coupler number after pre-training,the reason for the degradation of the feature extraction layer model in the original algorithm is analyzed,and the residual network with attention mechanism is introduced to improve the CTPN algorithm,and the ablation experiment is designed.The results show that the detection accuracy of the improved algorithm is improved by 8%.On this basis,two mainstream text recognition algorithms are used to recognize the coupler number.A comparative experiment is designed to verify the recognition accuracy and recognition efficiency of the two algorithms.The results show that the average recognition accuracy of the algorithm based on CTC is 10% higher than that of the algorithm based on Seq2 Seq on the coupler number recognition dataset,and the average recognition time of the former is only 20% of the latter.(3)An experimental platform for coupler number recognition is built,and a coupler number recognition system is developed based on the above algorithms.The system has the functions of detection mark,recognition display and result statistics,and the recognition accuracy of coupler number on the experimental platform reaches 96%.