Recognition of Railway Wagon Number Based on Deep Learning Networks

The existing electronic label of railway train has provided scan and recognition of wagon number. However, electronic labels have risks of losing, copying, counterfeiting and blocking, so that electronic labels cannot become an authentic and reliable method for wagon number recognition. Therefore, this paper proposes an image recognition method for wagon number images to be electronic label auxiliary method. The proposed method consists of preprocessing, character detection and recognition using learning deep networks and post-processing. The preprocessing of image enhancement increases the contrast of wagon number images. Then, YOLOv3 is employed to detect and locate character objects. The detection results are input CNN to recognize character. The post-processing performs to fill missing character based on the text lines of wagon numbers. The experimental results indicate that although the character recognition rate reaches nearly 90%, the accuracy rate of text line recognition is undistinguished at 60.6%.