【作者】 马丽；

【导师】 侯文生；

【作者基本信息】 重庆大学 ， 生物医学工程， 2008， 硕士

【摘要】 人手依靠手指和手掌的配合能够完成各种精巧复杂的动作,在协调运动的时候,各个手指根据动作的不同发挥不同的作用。手指的力量和动作是反映手指协同运动,评价手部运动机能的重要参数。采用传统的传感器测量这些参数存在的缺陷是易受使用环境和场地的限制。由于手指活动受前臂肌肉控制,而肌电信号是反映神经肌肉的功能状态的重要指标,所以利用前臂的表面肌电(sEMG,surface electromyography)信号来间接评价手指的活动模式对运动科学,人机交互,假肢控制,人体工效学,康复训练等许多方面的研究具有重要的学术和实用意义。本文首先建立指力检测实验系统,用LabVIEW8.0设计指力反馈系统软件,将指力计检测的指力信息用USB数据采集卡采集到指力反馈系统软件中,给受试者提供力量反馈,指导受试者用力。然后在该系统上分别进行单个手指4N,6N,8N按压动作的实验;食指中指合力为10%,20%,30%最大随意收缩力量(maximum voluntary contraction,MVC)时,两指自然分配力量比例的实验;食指中指合力为30%MVC,力量分配比例为1:1、5:2的实验,采集实验过程中指浅屈肌(flex digitorum superficials ,FDS)、指伸肌(extensor digitorum ,ED)的表面肌电信号,接着对这些动作的sEMG信号进行20-500Hz的滤波,计算FDS,ED的sEMG信号的均方根(RMS), AR系数,C0复杂度,比较相同肌肉不同力量水平时的RMS,食指中指力量比例分配不同时的C0复杂度,通过概率型神经网络(PNN)和学习矢量量化(LVQ)神经网络对食指、中指同等力量水平下的动作进行了分类。通过对实验采集的肌电信号的分析,得到以下一些结果:(1)食指、中指分别在4N、6N、8N三种力量水平下动作,其FDS、ED的肌电信号的RMS值随着力量的增加而递增。食指用力时FDS、ED的肌电信号的C0值大于同等力量下中指用力时的C0值;(2)以C0或AR系数为肌电信号的特征值用LVQ网络对同等力量水平下食指中指动作进行分类,准确率在80%以上;(3)以C0或AR系数为肌电信号的特征值用PNN网络分别对同等力量水平下食指中指动作进行分类,除了对4N力量下食指中指动作进行分类的准确率低于80%以外,其余的正确率均在80%以上。(4)对单个个体所有力量水平下的动作进行分类,以RMS和C0复杂度或者RMS和AR系数作为sEMG信号的特征值,采用PNN神经网络,分类正确率最高为95%。(5)食指中指自然力量分配下,合力达10%,20%,30%MVC时,表现为归一化RMS(%RMS)随着力量的增大而增加。(6)受试者食指中指合力为30%MVC,食指中指力量分配比例为1:1,5:2两种情况下,FDS、ED的RMS,C0值,虽然力量分配比例不同,但RMS,C0值无明显差异。研究结果表明,sEMG信号与手指动作具有相关性,RMS,AR模型系数,C0复杂度可以作为sEMG信号的特征值,反映肌肉的活动水平,估计手指的活动模式。随着单个手指(食指、中指)力量的增大,或者食指中指合力的增大,肌电信号的RMS值递增。以AR系数或C0复杂度作为指浅屈肌、指伸肌表面肌电信号的特征值,使用LVQ或PNN神经网络可以识别出单指特定力量下动作时是食指在用力还是中指在用力。给食指、中指人为分配不同的力量比例达到某一合力,这种情况下FDS,ED两块肌肉肌电信号的RMS、C0值均未表现出食指中指单独用力时所表现出的差异,同时,食指中指在两种给定的力量分配下,其值没有明显差异。更多还原

【Abstract】 Fingers and palm could concert to make lots of motion which is skillful and complex, different fingers had different functions according to the variety of the motion. Finger force was an important parameter to reflect finger synergetic motion and evaluate hand movement function. The restriction of environment and situation was the defects of the conventional sensors which had been used to gather the force parameter. Finger motion was controlled by the muscles of forearm, while the sEMG signal reflected the status of the never and muscle, so there might be some relationship between sEMG signal and finger motion, the sEMG (surface electromyography) signal could be used to estimate the motion mode of fingers, and it would play an important role in the research of mechanics, prosthetic hand, and rehabilitation training.First, an experiment system using for detecting finger force had been established. The LabVIEW-based finger force feedback software was designed, which could display the results of real-time acquisition of finger force and force track. Then, three experiments were conducted: (1) the experiment of single finger pressing at 4N, 6N, 8N; (2) the experiment of index and middle finger combined pressing at 10%, 20%, 30% MVC (maximal voluntary contraction); (3) the experiment of index and middle finger combined pressing force at 30%MVC, and index, middle finger force rate at 1:1 and 5:2. And the sEMG signal of FDS (flex digitorum superficials) and ED (extensor digitorum) was recorded. Next, the sEMG signal was filtered, and RMS (root mean square), AR model coefficient, C0 complexity as the characteristic of sEMG signal was calculated. After that, the RMS value of sEMG signal at different force had been compared, so did the C0 complexity in different force proportion. The motion of index finger or middle finger which pressed at a same force had been classified by the use of PNN (probabilistic neural networks) and LVQ (learning vector quantization) neural network. Some result could be obtained by analyzing the sEMG signal: (1) RMS value of both FDS and ED’s sEMG signal increased with index or middle finger force increased. When index finger pressed, the C0 value of FDS and ED’s sEMG signal was larger than middle finger pressed. (2). Using C0 value or AR coefficient as the input of the LVQ or PNN neural network could classify the motion between index finger and middle finger, and the correct rate was above 80%. (3). Using C0 value or AR coefficient as the input of the PNN neural network could classify the motion between index finger and middle finger, and the correct rate was above 80% except the motion at 4N. (4). For a single subject, The maximum correct recognition rate of all finger force level movement by using RMS and C0 value, or RMS and AR coefficient as the input parameter of PNN neural network could be 95%. (5). RMS value of both FDS and ED increased with index and middle finger combined force increased. (6). The RMS and C0 value didn’t showed differential when index and middle finger combined force at 30%MVC, and index, middle finger force rate at 1:1 and 5:2.The results of experiment showed that, a correlation between finger motion and sEMG signal had been represented. As the characteristic of sEMG signal, the RMS value, AR coefficient, and C0 complexity could reflect muscle active level, and could estimate finger motion. When single finger force increased or multi-finger combined force increased, the RMS of sEMG signal would raise. Using C0 value or AR coefficient as the input of the LVQ or PNN neural network could classify the motion between index finger and middle finger. The difference of sEMG signal didn’t represent when index and middle finger combined force at 30%MVC, and index, middle finger force rate at 1:1 and 5:2.更多还原