A Research Of Container Character Recognition Based On Neural Network

With the development of information and computer technology,people pay more and more attention to the container number identification technolog,in order to make the container transportation management more convenient. And the container number identification technology is already used in railway station、dock and so on. For example, when container trucks pass the door with container inspection system,the system can automatically recognize the container number, and put the result of identification system feedback to computer system. Thus,it can realize the dynamic tracking and management,and improve the efficiency of container transportation.Although the number identification system has many advantages, it also has some disadvantages. For example, the use of the system is affected by rain、fog、oil、 concavo-convex creases of container’s surface and other factors. So that the container number characters will distort or miss. These situations will affect the identification quality.The key of the container number identification mainly consists of four parts, including number positioning、number character segment、feature extraction and number character recognition. The thesis makes the further research on the four technologies, and gets some beneficial results.In the number positioning, according to the prior knowledge, the image must be preprocessed. Image preprocessing includes many contents, and the thesis mainly introduces the shades of gray image、image enhance processing、the binary、filtering、edge detection、mathematical morphology and other aspects of the processing. Finally, the image of number positioning is got. In the number character segment, in order to segmentation effectively, the thesis uses the vertical projection method for image segmentation eleven characters, and normalizes the division characters.In the number division part,in order to segment characters effectively,this paper uses vertical projection method to segment characters. For existing adhesion characters,this paper uses projection of two value method to segment characters. In the feature extraction part, many methods can be applied. Per-pixel feature extraction method、skeleton feature extraction method、13points of feature extraction method、the contour extraction method, and so on. Through the comparison of the several methods, the improvement of rough network feature extraction method is applied to extract character characteristic vector in the thesis finally. This kind of method can get local features and global features, and these can reflect the characteristics of every character very well.In the character recognition, the thesis uses the improved BP neural network to identify the characters. In order to avoid the confusion of similar characters,the thesis designs two BP neural network of three layers, and then get the results.The thesis makes the basic function of the algorithm under Matlab platform. The simulation gives the expected result.