基于TensorFlow平台的喷码字符日期码识别研究  

Research on inkjet character date code recognition based on TensorFlow platform

作　　者：阎晨阳 罗晓曙[1] 何富运[1] 马玲 YAN Chenyang;LUO Xiaoshu;HE Fuyun;MA Ling(Guangxi Normal University,Guilin 541004,China)

机构地区：[1]广西师范大学

出　　处：《现代电子技术》2019年第24期84-87,共4页Modern Electronics Technique

基　　金：广西科技重大专项资助项目(AA18118004)~~

摘　　要：针对传统的图像识别方法在喷码字符日期码识别中速度较慢的问题,文中使用TensorFlow平台,搭建一个卷积神经网络模型,实现端到端的喷码字符日期码识别。首先,对采集到的图像进行预处理,然后通过检测轮廓的方法提取图像中的喷码字符日期区域并进行尺寸归一化处理,再用训练好的卷积神经网络模型对图像进行端到端的识别。实验结果表明,此模型识别一张图像的平均时间为25 ms,对测试集的365张图像进行识别的准确率在99%以上。In allusion to the slow speed of traditional image recognition methods for inkjet character date code recognition,a CNN(convolutional neural network)model is built based on the TensorFlow platform to realize end-to-end recognition of inkjet character date codes. Firstly,the acquired image is preprocessed,and then the date code region of the inkjet character in the image is extracted by the outer contour detection method,by which the size normalization processing is carried out. The end-toend recognition is conducted for the image by means of the trained CNN model. The experimental results show that the model′ s average recognition time for an image is 25 ms,and the recognition accuracy of 365 pictures in the test set is over 99%.

关 键 词：日期码识别 喷码字符 TensorFlow平台 端到端识别 图像处理 仿真实验 

分 类 号：TN911.74-34[电子电信—通信与信息系统] TP391.4[电子电信—信息与通信工程]