摘要
随着人工智能的发展,计算机对于输入的手写字符识别需求越来越大,采用改进的卷积神经网络对手写字符进行识别分类。用VGGNet16模型构造卷积神经网络模型,每一层都加上批标准化,通过平均值池化对卷积层进行下采样,利用PRELU激活函数代替ReLU激活函数,最后通过Softmax分类器对手写字符图像进行分类。在MNIST手写数字数据集和EMNIST-bymerge手写字母及数字数据集下进行实验,改进的卷积神经网络模型在MNIST数据集中识别准确率提升到99.65%,在EMNIST数据集中识别准确率为90.37%。因此,改进模型识别准确率较高,适用于手写字符识别。
With the development of artificial intelligence,the demand for handwritten character recognition by computer is increasing.This paper uses the improved convolutional neural network to identify and classify handwritten characters.Firstly,the convolutional neural network model is constructed by VGGNet16 model.Each layer is batch-normalized,we use average pooling to downsample the convolutional layer,and the PRELU activation function is used to replace the ReLU activation function.Finally,the handwritten char⁃acter image is classified by Softmax classifier.Experiments were carried out under the MNIST handwritten digital dataset and the EM⁃NIST-bymerge handwritten character and digital dataset dataset.The improved convolutional neural network model improved the recog⁃nition accuracy in the MNIST dataset to 99.65%,and the recognition accuracy in the EMNIST dataset reached 90.37%.Therefore,the improved model has a higher recognition accuracy and is suitable for handwritten character recognition.
作者
丁蒙
戴曙光
于恒
DING Meng;DAI Shu-guang;YU Heng(School of Optical-Electrical and Computer Engineering,University of ShangHai for Science and Technology,Shanghai 20093,China)
出处
《软件导刊》
2020年第1期275-279,共5页
Software Guide
关键词
人工智能
卷积神经网络
手写字符识别
全局平均池化
artificial intelligence
convolutional neural network
handwritten character recognition
global average pooling