The Car Number Recognition Of Railway Truck Based On Machine Learning

The recognition of car number is an important area of artificial intelligence.The aim is to obtain the number information in the image by corresponding processing,calculation and feature extraction,so recognizing the car number.With the rapid revolution of railway transportation and the rapid growth of freight volume,the recognition of truck number has been widely used in location tracking and abnormal state detection of truck,it is an essential basis for the safe transportation of truck.Unlike the characteristics of the car number,the truck number has no fixed position,color and font,and it may be intermittent when it is painted.At this time,using traditional image processing technology for the recognition of truck number is often difficult to achieve the performance requirements.At present,there are few studies on machine learning to solve the difficulty of intermittent truck numbers and different truck models.The effect of truck number recognition is improved in the thesis by the machine learning method.Aiming at the problem that the existing recognition algorithm of truck number has low accuracy and weak robustness,in this thesis,the machine learning method is used to realize the recognition of the truck number.The main works are listed as follows:(1)Study the several types of technologies and algorithms in the existing truck number recognition module.(2)In the detection of truck number areas,analyze and study the method based on image processing,SSD(Single Shot MultiBox Detector)and Faster RCNN.The detection method based on Faster RCNN in this paper has high precision.What's more,analyze and explain the experimental results.(3)In the aspect of truck number recognition,the method of HOG(Histogram of Oriented Gradient)feature combined with the support vector machine is studied,but it has poor performance in recognition,and it can't effectively solve the problem of discontinuous truck number.So,this thesis proposes a new idea,which uses a single character and half a character as the training data set to design a convolutional neural network,it realizes the dynamic segmentation recognition of the truck number,and improves the recognition rate on the discontinuous truck number.(4)Thousands of image data sets are manually tagged based on the acquired truck image.Then,analyze and compare the performance of the proposed methods and the existing algorithms through multiple sets of experiments.(5)Summarize and reflect on the work of this paper,also analyze and forecast the next work in future.