计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (15): 178-185.DOI: 10.3778/j.issn.1002-8331.2004-0254

• 模式识别与人工智能 • 上一篇    下一篇

集装箱箱号字符识别算法研究

曹林根,宓超   

  1. 上海海事大学 物流工程学院,上海 201306
  • 出版日期:2021-08-01 发布日期:2021-07-26

Research on Recognition Algorithms for Container Code Character

CAO Lingen, MI Chao   

  1. College of Logistics Engineering, Shanghai Maritime University, Shanghai 201306, China
  • Online:2021-08-01 Published:2021-07-26

摘要:

针对集装箱箱号图像中存在的光照不均、箱号的偏转和倾斜等因素,着重研究箱号字符识别中的关键技术问题。对于箱号图像光照不均问题,采用一种改进型的差分边缘检测粗定位算法;利用改进的最小二乘法有效地解决箱号偏转难以精确定位问题;运用基于梯度下降投影字符矫正及分割算法,实现对倾斜箱号的校正与分割;采用BP神经网络进行字符识别。对1?050幅不同条件的拍摄图像进行实验,结果表明上述算法相对于传统算法与深度学习算法,综合识别率明显提高,且符合实时性要求。

关键词: 集装箱箱号识别, 差分边缘, 最小二乘法, 倾斜校正, BP神经网络

Abstract:

Aiming at the factors such as uneven illumination, deflection and tilt of container number in the image of container number, the key technical problems in character recognition of container number are mainly studied. For the problem of uneven illumination of container number image, an improved rough location method of differential edge detection is adopted. The improved least squares method is used to effectively solve the problem of container number deflection which is difficult to locate accurately. A gradient descending projection character correction and segmentation algorithm is used to realize the correction and segmentation of tilted container number. The BP network is used for character recognition. Experiments on 1?050 images captured under different conditions show that the comprehensive recognition rate of the above algorithm is significantly improved compared with traditional algorithm and depth learning algorithm, and it meets the real-time requirements.

Key words: container code recognition, differential edge, least square method, tilt correction, BP neural network