基于双迭代聚能量字典学习的数据压缩算法  

Data Compression Algorithm Based on Dual-iteration Concentrated Dictionary Learning

作　　者：代少飞 刘文波[1,2] 王郑毅 李开宇 DAI Shaofei;LIU Wenbo;WANG Zhengyi;LI Kaiyu(College of Automation Engineering,Nanjing University of Aeronautics&Astronautics,Nanjing 211106,China;Non-Destructive Testing and Monitoring Technology for High-Speed Transport Facilities Key Laboratory of Ministry of Industry and Information Technology,Nanjing 211106,China)

机构地区：[1]南京航空航天大学自动化学院,南京211106 [2]高速载运设施的无损检测监控技术工业和信息化部重点实验室,南京211106

出　　处：《数据采集与处理》2021年第6期1147-1156,共10页Journal of Data Acquisition and Processing

基　　金：国家重点研发计划(2018YFB2003304)资助项目;国家自然科学基金(61871218)资助项目;中央高校基本科研业务费(NJ2019007,NJ2020014)资助项目。

摘　　要：针对基于稀疏表示(Sparse representation,SR)的数据压缩压缩率低、重构精度低等问题,本文提出一种基于双迭代的聚能量字典学习算法,把高维信号映射到低维特征空间,当低维特征空间保留高维原始信号越多的特征时,高维信号从低维特征空间中恢复出来的精度越高。为了使低维字典保留高维字典更多的主成分,本文提出了一个新的变换,被命名为?变换,能提升高维字典的能量集中性。除此之外,针对高维字典与低维字典的耦合关系,建立了双循环迭代训练,增加字典的能量集中性与字典的表达能力。实验表明,相比于传统算法,本文提出算法字典学习收敛速度提升了3倍以上。此外,该方法可以得到较高的压缩比和更高质量的重构信号。As the data compression methods based on sparse representation(SR)have the problems of low compression ratio and reconstruction accuracy, a dual-iteration concentrated dictionary learning algorithm is proposed. This algorithm maps high-dimensional signals to low-dimensional feature spaces. If features of the high-dimensional original signal are retained by the low-dimensional feature space,higher accuracy will be achieved when the high-dimensional signal is reconstructed from the low-dimensional feature space. To keep more principal components of high-dimensional dictionaries in low-dimensional dictionaries,a new transformation algorithm named ? transformation is proposed. It can improve the energy concentration of the high-dimensional dictionary. Further,aiming at the coupling relationship between the high-dimensional dictionary and the low-dimensional dictionary,a dual-iteration training method is established to improve the energy concentration and the expressive ability of the dictionary.Experiments show that,compared with the traditional algorithms,the convergence speed of the proposed algorithm is improved by more than three times. In addition,a higher compression ratio and a higher quality reconstructed signal are obtained.

关 键 词：字典学习 聚能量字典 稀疏表示 数据压缩 低维特征 

分 类 号：TN911[电子电信—通信与信息系统]