摘要
铝轮毂背腔字符分辨率较低、背景噪声较大,对其进行识别时不易提取几何特征和纹理特征。为此,提出一种基于改进卷积神经网络(CNN)的字符识别方法。在原始CNN的基础上引入改进的inception结构对网络构架进行优化,以提升计算资源的利用率,并在保持网络计算资源不变的前提下增加网络的宽度和深度,降低字符识别时间。实验结果表明,该方法训练准确率达99%以上,识别准确率达98.5%,识别效果优于支持向量机、BP神经网络等方法。
It is difficult to extract geometric and texture features when recognizing the characters in the aluminium wheel back cavity because of its low resolution and strong background noise.Therefore,a character recognition method based on improved Convolution Neural Network(CNN) is proposed.On the basis of the original CNN,an improved inception structure is introduced to optimize the network architecture to improve the utilization of computing resources,increase the width and depth of the network and reduce the time of character recognition while keeping the network computing resources unchanged.Experimental results show that the training accuracy of this method is over 99 % and the recognition accuracy is 98.5 %,which is better than that of Support Vector Machine(SVM) and BP neural network.
作者
程淑红
周斌
CHENG Shuhong;ZHOU Bin(School of Electrical Engineering,Yanshan University,Qinhuangdao,Hebei 066000,China)
出处
《计算机工程》
CAS
CSCD
北大核心
2019年第5期182-186,共5页
Computer Engineering
基金
国家自然科学基金(61601400)
河北省博士后择优项目(B2016003027)
关键词
卷积神经网络
inception结构
网络构架
背腔字符
损失函数优化
Convolution Neural Network(CNN)
inception structure
network architecture
back cavity character
optimization of loss function
