Research Of Locomotive Number Automatic Acquisition And Recognition System Based On Image Processing

With the rapid development of railway in China,locomotive's security issues are increasingly paid attention to.As a key component of locomotive on-line detection system,it is very important for locomotive number recognition system to achieve automation and intelligentialize.In the thesis,the recognition of locomotive number was studied.By using digital image processing and optoelectronic measurement,a locomotive number automatic acquisition and recognition system was developed based on image processing for dynamic on-line recognition of locomotive number.For locomotive number automatic acquisition and recognition system,there are three aspects were studied on classification of multiple locomotive types,localization of locomotive number,segmentation and recognition of locomotive number.The main work included is as follows:(1)Classification of multiple locomotive types.In recent years,deep learning becomes a hot research topic both in academia and applications.Aiming at recognition of multiple locomotive numbers under complex illumination,a new method based on Convolutional Neural Network(CNN)is proposed,and the result of parameter adjustment is analyzed.Experiments show that CNN classification model can perform better in classification of multiple locomotive types.And a new direction is provided for recognition of multiple locomotive numbers.(2)Localization of locomotive number.In the preprocessing step,a new method,combined result of locomotive type classification and extracting the corresponding color channel,is proposed for gray scale transformation.It increases the contrast between locomotive number and background.In the threshold segmentation step,aiming at non-uniform illumination and variational illumination,a method combined local gray characteristics is proposed to realize threshold segmentation of locomotive number.Finally using morphological processing and prior knowledge to achieve the ultimate locomotive number positioning.This algorithm is more insensitive to the localization of multiple locomotive numbers under complex illumination,and performs better in binarization and robustness.(3)Segmentation and recognition of locomotive number.In the character segmentation step,we use minimum bounding rectangle information of contour to realize segmentation of locomotive number character,and eliminate the breaks of characters due to painting by using morphological closing operation.According to the particularity of Chinese characters,regarding them as a whole region to achieve segmentation and acquisition,this method decreases segmentation error due to Chinese character stroke separation.In the character recognition step,a method combined Histogram of Oriented Gradient(HOG)feature and Support Vector Machine(SVM)is proposed to realize multiple locomotive number characters recognition.Experimental results show that compared with the traditional recognition algorithm,this method has a high recognition rate and a strong robustness.The locomotive number automatic acquisition and recognition system based on the above-mentioned technology has been tested online,and the accuracy of multiple locomotive numbers recognition has been reached 96.81%,which lays a good foundation for the production of locomotive number automatic acquisition and recognition system.