手指肌电信号稀疏分解重构与活动段特征提取研究

针对传统信号处理方法在非平稳信号处理中的局限性问题,对稀疏分解思想和自适应过完备原子库进行了研究,提出了将稀疏分解思想应用到表面肌电信号处理中的方法。采用数据分割的方式,对原始信号进行了预处理。在正交匹配追踪算法的基础上,利用K均值-奇异值分解(K-SVD)算法构造了自适应过完备原子库,对分割后的各个样本块分别进行了稀疏分解,将其多维特征重构为一维稀疏系数。同时,以便于实际应用与连续控制为原则,对每个样本块的稀疏系数进行了重组,用单个特征值表征了样本块的多维特征。数据分析结果表明,重构后的一维稀疏系数可以保留四维原始信号的绝大部分能量,而重组后的特征值可以准确反映原始信号活动段的变化。

Sparse decomposition and reconstruction of finger EMG and feature extraction of active segment

Aiming at the limitation of traditional signal processing method in non-stable signal processing, Sparse Decomposition and self-adaptive overcomplete dictionary were studied. Sparse Decomposition was investigated in s EMG processing. And data partitioning was used in the signal preprocessing. Based on Orthogonal Matching Pursuit, the multi-dimensional characteristic of sample blocks was reconstructed to one-dimensional sparse coefficient by self-adaptive overcomplete dictionary. And the dictionary was built by K-SVD. Meanwhile, in order to facilitate practical application and continuous control, sample block sparse coefficients were recombined into single eigenvalue. The multi-dimensional characteristic of sample block was shown in this signal eigenvalue. The result of data analysis indicates that one-dimensional sparse coefficient inherits most energy from the original four-dimensional signal; the changes of original signal active segment could be reflected accurately by single eigenvalue. [ABSTRACT FROM AUTHOR]