基于块稀疏贝叶斯学习压缩感知的心音重构  

Reconstruction of heart sound based on CS and BSBL

作　　者：甘凤萍[1] 王海滨[1] 房玉 张凯[1] 秦国瑾 赵逍 

机构地区：[1]西华大学信号与信息处理重点实验室,四川成都610039 [2]山口大学理工学研究科

出　　处：《计算机工程与设计》2016年第4期1037-1041,共5页Computer Engineering and Design

基　　金：四川省科技厅支撑基金项目(2012GZ0019);四川省重点实验室开放研究基金项目(szjj2013-014);2015年西华大学研究生创新基金项目;西华大学2015年研究生创新基金项目(ycjj2015099)

摘　　要：为提高体域网远程传输心音信号的重构精度、运行时间及处理数据量,对一种基于块稀疏贝叶斯学习的压缩感知重构心音方法进行研究。在传感节点端对心音信号做分块处理,进行离散余弦变换字典训练;通过稀疏二进制矩阵对心音信号进行压缩,并传送至终端;利用块稀疏贝叶斯学习对终端压缩的心音重构,将重构结果与传统的正交匹配追踪结果比较。实验结果表明,块稀疏贝叶斯学习算法比正交匹配追踪算法重构的结构相似度高0.2-0.3,在信噪比方面高10db-30db,所提方法具有重构精度高,处理心音数据量大,运行时间快的显著优势。To improve the reconstruction accuracy,decrease the run-time and increase the processing data amount of body area network for the heart sound signals remote transmission,a heart sound compressed sensing reconstruction method based on the block sparse Bayesian learning（BSBL）was discussed.The block processing and discrete cosine transform（DCT）dictionary training of heart sound signal were applied at the sensor nodes.The heart sound signal was compressed via a sparse binary matrix and was transmitted to the terminal.The BSBL was used to reconstruct the compressed heart sound signal in terminal and the reconstructed result was compared with that of the conventional orthogonal matching pursuit（OMP）.Experimental results show that the structural similarity of BSBL is 0.2-0.3higher than that of the OMP,and signal to noise ratio of BSBL is 10db-30 db higher than that of OMP.The proposed compressed sensing method which can acquire high reconstruction accuracy,process large amount of heart sound data,and run fast,has obvious practical value for the future research.

关 键 词：压缩感知 块稀疏贝叶斯学习 正交匹配追踪 心音 体域网 

分 类 号：TN911.72[电子电信—通信与信息系统]