提出了联合多层次深度特征的合成孔径雷达(SAR)目标识别方法。采用卷积神经网络(CNN)学习SAR图像的多层次深度特征。多层次的深度特征从不同方面描述原始SAR图像中的目标特性,从而为目标识别提供更充分的决策依据。为了充分发掘不同层...提出了联合多层次深度特征的合成孔径雷达(SAR)目标识别方法。采用卷积神经网络(CNN)学习SAR图像的多层次深度特征。多层次的深度特征从不同方面描述原始SAR图像中的目标特性,从而为目标识别提供更充分的决策依据。为了充分发掘不同层次深度特征的独立特性以及它们之间的内在关联,采用联合稀疏表示对多层次的深度特征进行联合分类。根据各层次特征的整体重构误差判定目标类别。采用MSTAR (Moving and Stationary Target Acquisition and Recognition)公共数据集对提出方法进行了性能测试。实验结果表明,该方法的识别性能显著优于现有的SAR目标识别方法。展开更多
针对深度学习应用技术进行了研究性综述。详细阐述了RBM(受限玻尔兹曼机)逐层预训练后再用BP(反向传播)微调的深度学习贪婪层训练方法,对比分析了BP算法中三种梯度下降的方式,建议在线学习系统采用随机梯度下降,静态离线学习系统采用随...针对深度学习应用技术进行了研究性综述。详细阐述了RBM(受限玻尔兹曼机)逐层预训练后再用BP(反向传播)微调的深度学习贪婪层训练方法,对比分析了BP算法中三种梯度下降的方式,建议在线学习系统采用随机梯度下降,静态离线学习系统采用随机小批量梯度下降;归纳总结了深度学习深层结构特征,并推荐了目前最受欢迎的五层深度网络结构设计方法。分析了前馈神经网络非线性激活函数的必要性及常用的激活函数优点,并推荐Re LU(rectified linear units)激活函数。最后简要概括了深度卷积神经网络、深度递归神经网络、长短期记忆网络等新型深度网络的特点及应用场景,归纳总结了当前深度学习可能的发展方向。展开更多
针对隐式数据单纯利用隐反馈信息往往难以获取较好推荐性能的问题,提出一种融合元数据及隐式反馈信息的多层次深度联合学习(multi-level deep joint learning,MDJL)推荐方法。它利用双深度神经网络共同学习,其中一个网络利用隐式反馈学...针对隐式数据单纯利用隐反馈信息往往难以获取较好推荐性能的问题,提出一种融合元数据及隐式反馈信息的多层次深度联合学习(multi-level deep joint learning,MDJL)推荐方法。它利用双深度神经网络共同学习,其中一个网络利用隐式反馈学习用户及项目个体个性化关系,另一个网络利用元数据学习高层次群体共性化关系,从而有效地表达用户偏好,使MDJL框架在个体及群体因素间达到平衡。最后,MDJL推荐算法在Movie Lens 100K和MovieLens 1M两个公开数据集上进行实验评估。结果表明,该算法比其他基线方法表现出了更为优越的推荐性能。展开更多
Automatic signature generation approaches have been widely applied in recent traffic classification.However,they are not suitable for LightWeight Deep Packet Inspection(LW_DPI) since their generated signatures are mat...Automatic signature generation approaches have been widely applied in recent traffic classification.However,they are not suitable for LightWeight Deep Packet Inspection(LW_DPI) since their generated signatures are matched through a search of the entire application data.On the basis of LW_DPI schemes,we present two Hierarchical Clustering(HC) algorithms:HC_TCP and HC_UDP,which can generate byte signatures from TCP and UDP packet payloads respectively.In particular,HC_TCP and HC_ UDP can extract the positions of byte signatures in packet payloads.Further,in order to deal with the case in which byte signatures cannot be derived,we develop an algorithm for generating bit signatures.Compared with the LASER algorithm and Suffix Tree(ST)-based algorithm,the proposed algorithms are better in terms of both classification accuracy and speed.Moreover,the experimental results indicate that,as long as the application-protocol header exists,it is possible to automatically derive reliable and accurate signatures combined with their positions in packet payloads.展开更多
文摘提出了联合多层次深度特征的合成孔径雷达(SAR)目标识别方法。采用卷积神经网络(CNN)学习SAR图像的多层次深度特征。多层次的深度特征从不同方面描述原始SAR图像中的目标特性,从而为目标识别提供更充分的决策依据。为了充分发掘不同层次深度特征的独立特性以及它们之间的内在关联,采用联合稀疏表示对多层次的深度特征进行联合分类。根据各层次特征的整体重构误差判定目标类别。采用MSTAR (Moving and Stationary Target Acquisition and Recognition)公共数据集对提出方法进行了性能测试。实验结果表明,该方法的识别性能显著优于现有的SAR目标识别方法。
文摘针对深度学习应用技术进行了研究性综述。详细阐述了RBM(受限玻尔兹曼机)逐层预训练后再用BP(反向传播)微调的深度学习贪婪层训练方法,对比分析了BP算法中三种梯度下降的方式,建议在线学习系统采用随机梯度下降,静态离线学习系统采用随机小批量梯度下降;归纳总结了深度学习深层结构特征,并推荐了目前最受欢迎的五层深度网络结构设计方法。分析了前馈神经网络非线性激活函数的必要性及常用的激活函数优点,并推荐Re LU(rectified linear units)激活函数。最后简要概括了深度卷积神经网络、深度递归神经网络、长短期记忆网络等新型深度网络的特点及应用场景,归纳总结了当前深度学习可能的发展方向。
基金supported by the National Key Basic Research Program of China (973 Program) under Grant No. 2011CB302605the National High Technical Research and Development Program of China (863 Program) underGrants No. 2010AA012504,No. 2011AA010705+1 种基金the National Natural Science Foundation of China under Grant No. 60903166the National Science and Technology Support Program under Grants No. 2012BAH37B00,No. 2012-BAH37B01
文摘Automatic signature generation approaches have been widely applied in recent traffic classification.However,they are not suitable for LightWeight Deep Packet Inspection(LW_DPI) since their generated signatures are matched through a search of the entire application data.On the basis of LW_DPI schemes,we present two Hierarchical Clustering(HC) algorithms:HC_TCP and HC_UDP,which can generate byte signatures from TCP and UDP packet payloads respectively.In particular,HC_TCP and HC_ UDP can extract the positions of byte signatures in packet payloads.Further,in order to deal with the case in which byte signatures cannot be derived,we develop an algorithm for generating bit signatures.Compared with the LASER algorithm and Suffix Tree(ST)-based algorithm,the proposed algorithms are better in terms of both classification accuracy and speed.Moreover,the experimental results indicate that,as long as the application-protocol header exists,it is possible to automatically derive reliable and accurate signatures combined with their positions in packet payloads.
