Recognition Technology Of Railway Freight Car Number Based On Machine Vision

Car number recognition is an important field of artificial intelligence.The aim is to obtain the number information in the image by processing,calculating and extracting the characteristics accordingly,so as to identify the number of the car.With the rapid development of railway passenger and freight transportation and the rapid growth of freight volume,truck number identification is widely used in truck location tracking and truck abnormality detection,which is an important basis for truck safety operation.Unlike the characteristics of license plates,license plate numbers do not have a fixed position,color and font,and can break when painting.At this time,traditional image processing technology for vehicle number recognition is often difficult to meet performance requirements.At present,there is little research on using machine learning methods to solve the problems of license plate number discontinuity and identification of different models.This article attempts to improve the effectiveness of machine learning-based vehicle number recognition.Aiming at the problems such as low accuracy,weak robustness and poor detection effect of the existing image-based truck number recognition algorithm,this paper uses machine learning-based method to realize the accurate identification of truck number.First of all,the images used in the experiment are classified into characters defaced and characters are relatively intact.For the character-perfect picture first using the Faster RCNN model,accurately detected the position of the number,not only overcome the interference of light,but also achieve the detection of different truck models,and then proposed based on convolutional neural network dynamic segmentation recognition method,solve the problem of number break,improve the accuracy of identification.For character-stained pictures,this paper uses image patching algorithm based on texture synthesis to fill the gap before identifying them.The main work of the paper is as follows:(1)In the area detection of vehicle number,the number detection based on image processing,the area detection of vehicle number based on SSD(Single Shot Multi Box Detector)and the number detection method based on Faster RCNN are compared,and the Method RCNN detection method adopted in this paper is described with high accuracy,and the experimental results are analyzed and explained.(2)In terms of car number character recognition,a model recognition method based on characteristics and support vector machine is compared,but the accuracy of symbol recognition is low,which can not effectively solve the problem of vehicle fragmentation.This paper is designed to use a single character and half the number of characters as a data set for training,design a convolutional neural network to achieve the dynamic segmentation and recognition of train numbers,so as to improve the recognition rate of discontinuous train numbers.(3)For the defaced car number,the image repair algorithm based on texture synthesis is used to fill the hole,which makes the whole recognition process more perfect and the application range wider.(4)The contents of the paper are summarized and considered,and the follow-up work is analyzed and looked forward to.