中文摘要：

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.

英文摘要：

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.