Automatic Recognition Of Railway Wagon Numbers Based On Images

The acquisition of railway wagon numbers is a basic task of railway transportation system. The traditional way is recording numbers manually, which is an error-prone task and has low efficiency. Existing automatic wagons identification methods require hand-crafted image features, whose efficiency is not high. In this paper, an automatic recognition method for wagon numbers has been built via digital image processing and deep learning technology.Firstly the image pre-processing technologies, including image enhancement and image de-noising technology were studied. By image pre-processing it can improve the image contrast and reduce image noises while preserving image edges. Secondly, to determine the location of the wagon numbers in the image, how to find the candidate character areas were studied. We used the color space and the key points density of the image synthetically to determine the character candidate areas. Then, a deep convolution neural network model was built for automatically extracting image features, and precise positioning of the wagon numbers in the image can be detected. Finally, we made a comparative study of several commonly used classifier:k-nearest neighbor classifier, Softmax classifier and support vector machine classifier. Experimental results showed that k-nearest neighbor classifier correctly lowest rate, Softmax classifier and support vector machines the rate is very correct, support vector machine is slightly higher. So the support vector machines was adapted to make the identification.The proposed method can identify wagons accurately, which indicate the proposed method can provide technical support for the operation and management of the railway system automatically.