卷积神经网络的研究进展综述  

Review of Convolution Neural Network

作　　者：杨斌[1] 钟金英 

机构地区：[1]南华大学电气工程学院,湖南衡阳421001

出　　处：《南华大学学报（自然科学版）》2016年第3期66-72,共7页Journal of University of South China：Science and Technology

基　　金：南华大学青年英才支持计划基金资助项目(聘字2014-004号);南华大学校内博士启动基金资助项目(2011XQD29);国家自然科学基金资助项目(61102108);湖南省优秀博士学位论文基金资助项目(YB2013B039)

摘　　要：深度学习(deep learning,DL)强大的建模和表征能力很好地解决了特征表达能力不足和维数灾难等模式识别方向的关键问题,受到各国学者的广泛关注.而仿生物视觉系统的卷积神经网络(convolutional neural network,CNN)是DL中最先成功的案例,其局部感受野、权值共享和降采样三个特点使之成为智能机器视觉领域的研究热点.对此,本文综述CNN最新研究成果,介绍其发展历程、最新理论模型及其在语音、图像和视频中的应用,并对CNN未来的发展潜力和发展方向进行了展望和总结.Deep learning theory has received extensive attention of scholars all over the world because of its powerful modeling and high representational abilities.It solved the key problems of pattern recognition, such as the insufficiency of expression ability and dimen- sionality curse. Convolutional neural network learning, which imitates the biological vision （CNN） is a successful component of deep system. Local receptive field, sharing weights and down sampling are three important characteristics of CNN which lead it to be the hots- pot in the field of intelligent machine vision.Therefore,this paper summarizes the latest re- search works of CNN. Firstly, the history of CNN is introduced. Secondly, state-of-the-art modified models of CNN are reviewed.Then ,the applications of CNN in speech ,image and video processing are illustrated.Finally, the development trends of CNN are concluded.

关 键 词：深度学习 卷积神经网络 特征提取 智能识别 

分 类 号：TP412.14[自动化与计算机技术]