Research On The Algorithms Of Bogie Recognition Based On Machine Vision

The bogie is one of the most important large parts of the high-speed EMU and the production of the bogie is also a key part of the entire high-speed train manufacturing.The machining center automatically selects different machining procedures according to bogie types.In order to avoid accidents,bogie must be identified before processing.At present,the identification of bogies mainly divided into manual identification and radio frequency identification.But manual identification is not conducive to automated production,the accuracy of radio frequency identification is low.Machine vision technology is developing rapidly in recently years.The use of machine vision technology for bogie identification can not only reduce manual intervention to achieve automatic production,improve the identification efficiency and accuracy,but also improve the degree of workshop information management,lay the foundation for building an intelligent manufacturing workshop in the future.This paper studies the bogie recognition algorithm from image matching algorithm and deep learning algorithm based on machine vision technology.The main research content of the paper is as follows:(1)Bogie image acquisition and preprocessing.According to the bogie's appearance features and placement scene,the hardware selection and construction of bogie image acquisition system are completed,and the bogie image acquisition is realized.In order to better extract features such as image contours,the image preprocessing algorithm for bogie is proposed from two aspects: image color space and noise reduction.(2)Research on the algorithm of geometric feature recognition for bogie based on image matching.To solve the problem that the existing recognition algorithm based on image matching has low accuracy in the field of bogie recognition,the color image segmentation algorithm is used to improve the algorithm,an image matching algorithm combined with color image segmentation is proposed.An adaptive segmentation algorithm based on HSV color space is used to segment the bogie image,and more complete edge features of the bogie image are used for image matching.By comparison,it is found that the image matching algorithm combined with the color image segmentation is more complete,the recognition accuracy is higher,the time is shorter and the anti-interference ability is stronger.(3)Research on the algorithm of bogie classification and recognition based on deep learning.In view of the complexity and poor generalization ability of artificial feature extractor,bogies with many categories but few samples and uneven distribution,a bogie recognition algorithm based on deep learning is proposed.Combining the advantages of Siamese network model and stacked network model,an unbalanced small sample recognition algorithm for bogie based on Siamese network is proposed,features are extracted by stacked network for classification and recognition under Siamese network.It is easy to see from the four comparison experiments that the accuracy of the method in this paper is 4.2% higher than stacked network on bogie data sets.At the same time,the method also solves the problem of model retraining caused by new categories and is more suitable for non-equilibrium small sample bogie classification and recognition.