摘要
为了提高识别效率并减少人工成本,采用深度学习的方法对生产日期图像进行识别。首先对生产日期图像进行预处理,使用水平投影分割算法并提出一种区域最大值分割的方法将图像中的干扰字符去除,只留下数字、字母和汉字字符。然后创建一个由生产日期图像中常包含的数字、英文、汉字字符所组成的可扩展的数据集。最后构建一个卷积神经网络模型并将数据集送入训练以获得较高的识别准确率。经测试基于卷积神经网络的识别方法对生产日期识别的准确率高达98%。
The work aims to improve recognition efficiency and reduce labor costs by using deep learning to recognize production date images.Firstly, the production date image is preprocessed, and the horizontal projection segmentation algorithm is used for segmentation. The region maximum segmentation method is proposed to remove the interfering characters in the image, leaving only numbers, letters and Chinese characters. Then create an extensible data set consisting of numbers, English, and Chinese characters that are often included in the production date image. Finally, a convolutional neural network model is constructed to train the data set to obtain higher recognition accuracy. The accuracy of the recognition method based on convolutional neural network on the production date is as high as 98%.
作者
胡蝶
侯俊
张全年
何金亭
王宗宜
Hu Die;Hou Jun;Zhang Quannian;He Jingting;Wang Zongyi(School of Optical Electrical and Computer Engineering,Shanghai University of Science and Technology,Shanghai 200215,China)
出处
《电子测量技术》
2020年第1期152-156,共5页
Electronic Measurement Technology
基金
上海自然科学基金(12ZR1420800)。
关键词
生产日期识别
卷积神经网络
区域最大值分割算法
投影分割算法
production date identification
convolutional neural network
regional maximum segmentation algorithm
projection segmentation algorithm
