第3 5 卷第1 2 期
2 0 1 9 年1 2 月
信号处理
J o u rn a l o f S ig n a l P ro c e s s in g
V o l.3 5 N o. 12
D ec . 2 0 1 9
文章编号:1003-0530(2019)12-2055-07
核卷积神经网络研究与应用
包志强赵志超吕少卿黄琼丹
( 西安邮电大学通信与信息工程学院,陕西西安710121)
摘 要 :利用核函数非线性映射的优势,结合卷积神经网络算法,提出一种基于核卷积神经网络^ 織 ‘ & ^ 。-
lu S n a l Neural N etw o rk, K e n e l-C N N ) 的新的网络学习模型。该方法首先对数据预处理,其次利用核卷积神经网络
对数据进行特征提取,最后,构建s ft m a x 分类器对数据进行分类。本网络将非线性映射引人卷积过程构成核卷
积过程,通过核卷积过程进一步增强模型的特征提取能力,在 M N IS T 手写数字库以及美国麻省理工学院提供的
M IT -B IH 心律失常数据库上实验验证,本文模型正确率分别为98. 5 % 、9 7 % ,均较好于卷积神经网络和支持向
量机,且本文模型具有较小的LO S S 值。
关键词:核函数;非线性映射;卷积神经网络;支持向量机
中图分类号:TP183 文献标识码:A D OI: 10. 1 6 7 9 8 /j. issn. 1003-053 0 . 20 1 9 .1 2 .01 4
引用格式:包志强,赵志超,吕少卿,等.核卷积神经网络研究与应用[J ]. 信号处理,2 0 1 9 , 3 5 (12 ): 20 55-2061.
DO I : 1 0 .1 6 7 9 8 /j. issn. 1 0 03-053 0 . 2019. 12. 014.
Reference form at: Bao Z h iqia n g,Zhao Zhich a o ,L li S haoqing,et al. Research and A p p licatio n of Kernel Convolutional
Neural Netw orks' J ] . Journal of Signal P ro c e s in g ,2 0 1 9 ,35 ( 12 ): 2 0 5 5 -2 0 6 1 .D O I: 10. 1 6 7 9 8 /j. is n . 10 0 3-0530.
2 0 1 9 .1 2 .0 1 4 .
Research and Application of Kernel Convolutional Neural Networks
A b s tr a c t: In this paper,using the advantages of kernel function nonlinear mapping,combined with convolutional neural
network algorithms,a newnetwork learning model kernel convolutional neural network (Kernel-CNN) is proposed. The
method first preprocesses the data,and secondly uses the kernel convolutional neural network to extract the features of the
data. Finally,the softma classifier is constructed to classify the data. This network introduces the nonlinear mapping into
the convolution process to form the kernel convolution process, and further enhances the feature extraction ability of the
model through the kernel convolution operation. Experimental verification was performed on the MNIST handwritten digital
library and the MIT-BIHarrhythmia database provided by the Massachusettt Institute of Technology. The correct rate of this
model is 98. 5% and 97%,respectively,which are better than convolutional neural networks and support vector machines.
The model also has a small LOSS value.
K e y w o rd s : kernel function; nonlinear mapping; convolutional neural network; support vector machine
近年来,卷积神经网络(C o n v o lu tio n a l N e u ra l N e t- 络作为深度学习中一种代表性的算法,具有出色的特
B a o Z h iq ia n g Z h a o Z h ic h a o L u S h a o q in g H u a n g Q io n g d a n
( School of Communication and Information Engineering, Xi ’ an University of
Posts and Telecommunications,Xi’an, Shaanxi 710121,China)
i 引言
w o r k ,C N N )成为计算机视觉领域最流行的技术,它在
图像分类问题中发挥着重要的作用[1]。卷积神经网
收稿日期! 2019-09-19;修回日期! 2019-12-10
基金项目:陕西省教育厅科研计划项目资助(17JK0703) ; 陕西省重点研发计划资助项目(2018GY-150) ; 西安市科技计划项目
(201805040YD18CG24-3,GXYD17. 5)
评论0