基于双线性模型的动作肌电信号用户无关识别研究

发布时间：2018-05-18 02:18

本文选题：肌电控制 + 手势识别　；参考：《中国生物医学工程学报》2016年05期

【摘要】：动作肌电信号具有个体差异性且不同动作的肌电信号是不同的,通过挖掘双线性模型的因素分解能力,将训练样本的特征矢量分解为用户相关和动作相关两大因素,通过确定因素的维度重构具有共性的训练样本特征。在测试样本特征重构阶段引入适应融合机制,更新模型参数重构测试样本特征。以11名受试者的4类动作为例,分别采用线性判别、K近邻分类算法和支持向量机,对比3种实验方案(多用户单天、单用户多天和基于双线性模型的多用户单天)的识别结果。实验表明,双线性模型的平均识别率最低为85%以上,相比于单纯的多用户单天识别结果(平均识别率不高于75%)有显著提高(P0.001),且相比于单用户多天的识别结果(平均识别率90%以上)差异性不显著(P0.24)。双线性模型为基于动作识别技术的非特定人肌电控制系统提供了交互方案,且该模型具备将多用户单天的数据看成单用户多天数据的能力,提供了用户训练负担降低的可行性。
[Abstract]:EMG signals have individual differences and EMG signals of different actions are different. By mining the factor decomposition ability of bilinear model, the feature vectors of training samples are decomposed into two major factors: user correlation and action correlation. By determining the factors of dimension reconstruction has the common characteristics of the training sample. The adaptive fusion mechanism is introduced in the phase of test sample feature reconstruction, and the model parameters are updated to reconstruct the test sample feature. Taking four kinds of actions of 11 subjects as examples, the recognition results of three experimental schemes (multi-user single day, single-user multi-day and bilinear model based multi-user single-day) were compared by using linear discriminant K-nearest neighbor classification algorithm and support vector machine. Experiments show that the average recognition rate of bilinear model is more than 85%. Compared with the simple multi-user recognition results (the average recognition rate is not higher than 7575), the P0.001D is significantly improved, and the difference is not significant compared with the single-user multi-day recognition results (the average recognition rate is more than 90%), and the difference is not significant (P 0.24%) compared with the single-user multi-day recognition results (the average recognition rate is more than 90%). The bilinear model provides an interactive scheme for an independent EMG control system based on motion recognition technology, and the model has the ability to treat multi-user single-day data as single-user multi-day data, and provides the feasibility of reducing the user training burden.
【作者单位】：合肥工业大学仪器科学与光电工程学院生物医学工程系;
【基金】：国家自然科学基金(61401138;81571760;61501164)
【分类号】：R318;TP391.41

【相似文献】