基于改进 LeNet-5 的牛奶生产日期识别研究-《计算机技术与发展》

文章信息/Info

Title:: Research on Milk Production Date Recognition Based on Improved LeNet-5

Author(s):: LEI Fei; SUN Kang; WANG Xue-li; Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China

Keywords:: production date; coding character; character recognition; LeNet-5; activation function

摘要:: 生产日期是牛奶包装上的强制标识之一,厂家需要在牛奶生产过程中对其进行检查,在检查合格后方可上市。当前国内对牛奶生产日期喷码字符的检测主要依靠人力,传统人工检测耗时长、成本高。为了解决上述问题,采集牛奶图像结合计算机视觉相关技术,提出了一种基于改进 LeNet-5 的牛奶生产日期字符识别方法。该方法主要是改变了传统LeNet-5 网络各层特征图的数量及大小等参数,减小其复杂程度,同时改进了激活函数,使用 ReLU 函数替代了最初的Sigmoid 函数,形成了适用于牛奶生产日期识别的新网络模型。实验结果表明,改进后的 LeNet-5 网络模型能够有效地识别牛奶生产日期,识别率达到 92.17% 。与传统 BP 神经网络识别算法相比,改进后的 LeNet-5 在提高牛奶生产日期识别率上有着明显的优势,并且识别速度较快。

Abstract:: The date of manufacture is one of the mandatory markings on the milk packaging. The manufacturer needs to inspect the milk during its production and check it before it can be marketed. At present,the detection of the code characters of the milk production date in China mainly relies on human resources,and the traditional manual detection takes a long time and costs a lot. In order to solve the above problems,based on milk image collection and computer vision related technology,we propose a milk production date character recognition method based on improved LeNet-5. This method mainly changes the parameters such as the number and size of the characteristic maps of the traditional LeNet-5 network,to reduce the complexity,and improves the activation function. The ReLU function is used to replace the original Sigmoid function,forming a new network model suitable for milk production date recognition. The experiment shows that the improved LeNet-5 model can effectively identify the milk production date,with the recognition rate of 92. 17% . Compared with the traditional BP neural network recognition algorithm,the improved LeNet-5 has obvious advantages in improving the milk production date recognition rate with faster recognition speed.