摘要
体表肌电信号会随着外部或人体内部环境变化而发生改变,这种时变特征使得固定参数肌电模式分类器的分类精度会随着时间的延长而下降。为了获得具有稳定性能的肌电假肢控制系统,在肌电模式分类器中加入自适应机制是很有必要的。本文以传统线性判别分析(Linear Discriminant Analysis,LDA)为基础,尝试在肌电模式分类器中引入三种自适应方案,并探讨了这三种方案在肌电模式分类应用中的优缺点。初步研究表明:自增强线性判别分析(Self-enhancing LDA,SELDA)分类器和循环训练集线性判别(Cycle Substitution LDA,CSLDA)分类器都能够将识别准确率提升5%左右。其中,SELDA是一种有效的自适应方案,而CSLDA可以得到更高的识别率提升和更好的稳定性,但是计算量较大,需要更大的代价。卡尔曼自适应线性判别(Kalman Adaptive LDA,KALDA)分类器单独使用效果不明显,需要进一步改进或结合其他方法使用。
When the surface electromyography (sEMG) signals change along with external or internal environment of the human body, general pattern classifiers will lead to a decrease of identification accuracy since they do not update their parameters adaptively. In order to adapt to the time-varying characteristics of sEMG signals, three kinds of adaptive algorithms for updating the parameters of a classifier during the use of artificial limb were introduced to improve the classification accuracy of time-variant sEMG signals. The pilot results of this study show that self-enhancing linear discriminant analysis is an effective solution and cycle substitution linear discriminant analysis presents the best performance but requires a large amount of calculations. The performance of the Kalman adaptive linear discriminant analysis is not prominent when it was used alone, and therefore it needs to be combined with other methods.
出处
《集成技术》
2013年第4期20-26,共7页
Journal of Integration Technology
基金
国家自然科学基金重点项目(61135004)
关键词
表面肌电信号
假肢控制
线性判别分析
自适应方法
sEMG signals
artificial limbs control
linear discriminative analysis
adaptive method
