基于神经网络的英语语义智能分析方法研究  

A study of English semantic intelligence analysis method based on neural network

作　　者：张慧[1] 许大炜[1] ZHANG Hui;XU Dawei(City College,Xi’an JiaoTong University,Xi’an 710018,China)

机构地区：[1]西安交通大学城市学院,陕西西安710018

出　　处：《电子设计工程》2020年第21期29-32,37,共5页Electronic Design Engineering

基　　金：陕西省社科基金(2018M30)。

摘　　要：针对英文语义智能化分析的应用需求,文中对英文字符的自动识别进行了研究。通过引入机器学习领域中的径向基函数(RBF)网络,提出了基于字符图像的英文字母识别方法。RBF使用高斯基函数替代传统神经网络中的激活函数,大幅提升了网络的比拟能力。RBF网络使用梯度下降法进行网络训练,增强网络的收敛性能,并提升训练速度。为了验证所提方法的性能,在开放数据集Englishhnd上进行了测试。结合实际的应用场景,测试着重关注RBF的抗噪声性能。仿真结果表明,在使用无噪声数据进行训练时,测试数据只有在噪声均值超过0.1后,该方法的识别错误率才会出现明显变化,较BP神经网络具有更强的抗噪性能。此外,该方法对于英文字符的识别精度可达到96.35%,AUC可达0.89,均优于BP神经网络。According to the application requirements of English Semantic Intelligent analysis,this paper studies the automatic recognition of English characters.By introducing the Radial Basis Function(RBF)network in the field of machine learning,an English character recognition method based on character image is proposed.RBF uses Gaussian basis function to replace the activation function in the traditional neural network,which greatly improves the network’s comparative ability;RBF network uses gradient descent method to train the network,which enhances the convergence performance of the network and improves the training speed.In order to verify the performance of the proposed method,it is tested on the open data set Englishhnd.Combined with the actual application scenario,the test focuses on the anti noise performance of RBF.The simulation results show that only when the noise mean value of the test data exceeds 0.1,the recognition error rate of this method will change significantly,which has stronger anti noise performance than BP neural network.In addition,the recognition accuracy of the method for English characters can reach 96.35%,AUC can reach 0.89,which are better than BP Neural Network.

关 键 词：语义分析 英文字符 神经网络 径向基函数 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]