The Method Of Recognizing Container Number Based On Convolutional Neural Networks

Digital image processing has many years of history, and many theories and algorithms have been developed for digital image processing. Container number recognition technology has a wide range of applications in the field of port transport. Currently, in Shanghai port, container number recognition technology is only applied on gate-in and gate-out with limited recognition efficiency. While in the yard operation and vessel handling, no container number recognition procedures are performed because of technology limit. With the popularity of deep learning and convolutional neural network(CNN) theory, it is possible for the container number recognition algorithm to become better and practical.After learning the relevant research results of convolutional neural network home and abroad, this thesis improves the container identification process, and applies it to the container identification technology. The thesis focuses on the following three parts:Firstly, this article introduces the basic knowledge of container, OCR equipment installation layout, analyzes the collected container number image, and finds out the distribution rule about letters and digits in container number. Then, it is image pre-processing stage. According to the distribution rule, the experimental comparison is researched to compare the effect of various gray transformation and filter to container number pictures. In this way, we can find out the most suitable gray algorithm and filter. Meanwhile, edge detection is done by Canny algorithm, to detect the useful information in image.In the second part, the corner comparative experiment is used to position container number. Each character image is segmented through vertical projection. As for the end of string, the elimination of frame is processed by a combined processing method, such as median filter, erosion, expansion and so on.Lastly, to further improve recognition rate of container number, convolutional neural network model is used to supervise training study of character image. Then, image recognition rate is compared with that of Google's open source library(tesseract – OCR). As a result, it proves that character processing, based on convolutional neural network, has a better effect.Through the Open CV function library and other tools on the X86 platform, the code of container number recognition is written and performed. Then the recognition rate rises up to 94.3%. Therefore, Convolutional neural network technology is so practical that it can be applied and generalized in the field of container number recognition during related terminal operation.