基于大腿残肢表面肌电信号的运动模式识别

发布时间：2018-08-05 20:19

【摘要】：穿戴假肢是下肢截肢者恢复日常生活的唯一途径,只有快速准确地识别残疾人的运动意图,才能实现对假肢的灵活控制。肌电信号能体现人体的运动意图,而且产生于人体真正运动之前,相比其它传感器信息具有显著优势。现阶段肌电信号运动模式识别的研究主要集中在健康人,但是残疾人由于肌肉的截断,其肌电信号与健康人存在很大的不同,因此利用残肢肌电信息识别不同运动模式具有重要的意义。本课题从残疾人出发,通过采集大腿残肢表面肌电信号识别下肢不同运动模式,识别结果可以为假肢提供控制源。主要的研究内容如下:1.分析截肢对下肢肌肉的影响,选定下肢肌群,确定肌电信号采集方案。针对现有采集方案的不足,为截肢者设计内嵌式传感器系统,采集不同运动模式下大腿残肢表面肌电信号。2.通过对不同运动模式肌电信号的分析,提出了一种基于移动窗的多数据段检测算法,实现了一个运动周期内多数据段的提取。并对多数据段进行特征提取,构建特征向量。3.针对误导和冗余特征值,利用基于遗传算法的BP神经网络降维、基于遗传算法的偏最小二乘法降维、基于平均影响值的神经网络特征降维的方法对特征值进行降维。根据降维结果,确定最优特征向量。经过降维,特征值维数明显下降,而且识别率有所提高。4.选定各数据段最优参数和数据段长度,用随机森林算法进行分段运动模式识别。根据识别结果,提出一种基于随机森林的二叉树周期性模式识别方法,实现了5种运动模式的周期性识别。实验结果证明,该方法提高了模式识别的准确率和实时性,有利于假肢的实时控制。
[Abstract]:Wearing artificial limbs is the only way to restore the daily life of the amputees of the lower extremities. Only by fast and accurate recognition of the movement intention of the disabled can the flexible control of the prosthesis be realized. The EMG signals can reflect the movement intention of the human body, and before the real movement of the human body, it has a significant advantage compared with its sensor information. The research of signal motion pattern recognition is mainly focused on healthy people, but the EMG signals of the disabled are very different from those of the healthy people because of the truncation of the muscles. Therefore, it is of great significance to identify different sports patterns by using the information of the myoelectric power of the disabled. The main research contents are as follows: 1. the main research contents are as follows: 1. analysis of the influence of amputation on the muscles of the lower limbs, the selection of the lower extremities and the acquisition scheme of EMG signal. The surface electromyography (EMG) signal.2. proposes a multi data segment detection algorithm based on the mobile window, which is based on the analysis of different motion patterns, and realizes the extraction of multiple data segments within a period of motion. The feature extraction is carried out for multiple data segments, and the feature vector.3. is constructed for the error and redundancy eigenvalues, and the BP based on genetic algorithm is used. Reducing dimension of neural network, reducing dimension based on partial least squares of genetic algorithm, reducing the dimension of eigenvalues based on eigenvalue reduction of neural network based on average influence value. According to the reduced dimension results, the optimal eigenvector is determined. After dimensionality reduction, the dimension of eigenvalues decreases obviously, and the recognition rate is improved to some extent of the optimal parameters and numbers of the selected data segments of.4.. According to the recognition results, a periodic pattern recognition method of two forked trees based on random forest is proposed to realize the periodic recognition of 5 kinds of motion patterns. The experimental results show that the method improves the accuracy and real time of pattern recognition, and is beneficial to the real time control of the prosthesis. System.
【学位授予单位】：河北工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：R496;TP391.4

【参考文献】