Research And Design Of Automatic Recognition System For Carriage ID Of Pelletizing Mineral Transportation Train

Automatic recognition system of railway freight train carriage number is an important part of railway informatization.It can increase productivity and be convenient for managing trains.Computer vision technology is widely noted due to its easy maintenance and low-cost relative to using RFID.Recognizing the carriage number of freight train by computer vision includes two steps which are character location and character recognition.In this dissertation,a deep learning method to overcome difficulties such as break and pollution in carriage ID character of pelletizing mineral transportation train is carried out and an automatic recognition system which can recognize railway freight train carriage numbers accurately is designed.Firstly,a simplified fully convolution network without up-sample layer is designed by transforming fully connected layer in convolution neural network which can distinguish character and background correctly to convolution layer.Character heat-map can be created by inputting source image into the fully convolution network which is designed in this dissertation.Carriage type text and carriage serial number can be extracted in high reliability area.Experiment results show that character location accuracy is 98.1%.After that a convolution neural network with multi-label classification ability is designed.Carriage type text is recognized by projection segmentation method combined with convolution neural network.Carriage serial number is recognized by putting all the numbers together before inputting into the multi-label convolution neural network because of its obvious break.Experiment results show that the method carried out in this dissertation can avoid misrecognition caused by break effectively and its accuracy can reach 99.84%.Finally,an automatic recognition system for carriage ID of pelletizing mineral transportation train is designed in this dissertation.The hardware part consists of a webcam,a router and a computer.The software part which consists of image processing implemented by OpenCV,selecting of character candidate area and character recognition implemented by Caffe is developed by VS2013 and GUI is programmed by Qt.Experiment results show that the automatic recognition system for carriage ID of pelletizing mineral transportation train designed in this dissertation is able to run stably and recognize carriage ID correctly.