Research On Related Algorithms Of Railway Freight Car Number Automatic Identification

This topic is based on image processing and pattern recognition of railway freight car number for identification. Because of the complexity of the railway freight car number image,in the process of practical application of railway freight car number, the accuracy of identification system based on image processing is always poor. This topic is research for the related algorithms of railway freight car number automatic identification system, which improves the localization rate and the recognition rate of railway freight car number. And it also can satisfy the actual needs.Firstly, the paper gives a deep research on the status and technique of the railway freight car number automatic identification system at home and abroad. On the basis of existing research, some related algorithms are improved. The whole system concludes four modules:image processing, locating car number area, character segmentation and character recognition.The paper main work is as follows:1. Image preprocessing, which mainly includes image gray processing, the image stretching, image smoothing.2. A novel car number localization algorithm based on line scanning and wavelet transform is introduced. The image edge detection is based on Sobel, then determines the upper and lower boundary of the car number area by scanning each row in the image, and gets the horizontal gray-level projection of image. At the same time, left and right boundaries of the image is determined by wavelet transform and setting adaptive threshold. Simulation results show that this plate localization algorithm has fast localization speed, high accuracy and robustness.3. An improved Ostu arithmetic and character segmentation is proposed. Image segmentation mainly includes binary and character segmentation. The improved Ostu is used in the process of binary, which is based on the algorithm of two-dimensional threshold segmentation. The improved Ostu arithmetic reduced the amount of calculation, and can achieve good effect. Character segmentation includes coarse segmentation and precise segmentation. Coarse segmentation is based on projection of horizontal and vertical. Coarse segmentation is aimed at the fracture character and adhesion character.4. An improved BP neural network is proposed. Due to the problem of low convergence in neural network, an improved BP neural network is proposed. The arithmetic is based on momentum of gradient descent method. The experimental results show that the algorithm can identify characters better and fast computing speed.