sEMG时频特征线性回归法与非线性神经网络法预测伸膝肌群极限功率保持能力测试中功率损失率的比较研究  

Linear Regression vs.Non-Linear Neural Network Predicts Power Loss Rate using s EMG Time-Frequency Features during the Retentive Capacity of Muscular Peak Power Test for Knee Extensor

作　　者：徐红旗[1] 史冀鹏[1] 张守伟[1] 徐欣[1] 杨传崎 杨帆[3] 张欣[4] 赵朝义[4] 

机构地区：[1]东北师范大学,吉林长春130024 [2]长春光华学院商学院,吉林长春130033 [3]北京航空航天大学生物医学工程学院,北京100191 [4]中国标准化研究院人类工效学标准化研究领域,北京100088

出　　处：《中国体育科技》2017年第2期53-63,共11页China Sport Science and Technology

基　　金：科技基础性工作专项(2013FY110200);中央高校基本科研业务费资助项目(14QNJJ032)

摘　　要：目的:拟比较s EMG时频特征线性回归法与非线性神经网络法预测伸膝肌群极限功率保持能力测试中功率损失率的差异。方法:BTE Primus^(RS)系统与肌电仪同步,40名男大学生膝关节重复性屈伸运动至疲劳,阻力设置50%等长峰值力矩,动作频率60次/min。求取每次伸膝阶段极限功率损失率(Power%),伸膝肌群sEMG时域(MAV%、RMS%)、频域(MNF%、MDF%)与瞬时频率(IMNF%、IMDF%)参数变化率,基于s EMG时频特征参数(MAV、ZC、SSC、WL)建立多层感知人工神经网络模型,求取功率真实值与估计值。结果:IMDF%能单独解释股内肌、股直肌与股外肌极限功率损失率的方差变异为6.33%、22.71%、12.31%,IMDF%联合其他时频参数一起能解释的方差变异为6.95%、25.93%和16.05%,非线性神经网络法求取的功率估计值能解释的方差变异为10.43%、34.23%和18.05%,且信噪比值逐步增大。线性与非线性技术功率真实值与估计值拟合所得两直线的斜率与截距有显著性差异(P<0.05)。结论:s EMG时频特征线性回归法与非线性神经网络法,均能很好地追踪人体神经肌肉系统动态工作疲劳过程中输出功率的损失,但后者的准确性要优于前者。Objective： This study compares a linear （linear regression） and a non-linear （neural net- work） power loss mapping using a set of features of the surface electromyogram recorded from the knee extensor during the retentive capacity of muscular peak power test. Methods： BTE Pri- musRS system and surface electromyography instrument were synchronized, 40 male college students repeated knee flexion and extension exercise to fatigue, with the load resistance for 50% of the isometric peak torque, and action frequency 60 times/min. The following variables were computed from each extension contraction： peak power loss （Power%）, and the changes rate of sEMG variables such as MAV%, RMS%, MNF%, MDF%, IMNF%, IMDF%. A multi-layer perceptron neural network was established, which consists of an interconnected group of artificial neu- rons （MAV, ZC, SSC, WL）, and used to calculate the actual and estimated changes in power. Results： IMDF% as a single parameter predictor accounted for 6.33%, 22.71%, 12.31% of the performance variance of changes in vastus medialis, rectus femorisa and vastus lateralis peak pow- er, IMDF% and other time-frequency parameters as a combination predictor accounted for 6.95%, 25.93%, 16.05%, the estimated changes of power were calculated by a non-linear neural network accounted for 10.43%, 34.23%, 18.05%, and the signal to noise ratios increased gradually. Two regression lines were fitted by the actual vs. the estimated changes in power output from both linear and non-linear techniques, and the slope and intercept of two lines had a statistical significance （P〈0.05）. Conclusion： Linear regression and non-linear neural network using sEMG time-frequency features, are two useful tools to map changes in muscular power loss during dynamic fatiguing task, but the latter＇s accuracy is better than the former.

关 键 词：极限功率 表面肌电 小波变换 瞬时频率 人工神经网络 

分 类 号：G804.63[文化科学—运动人体科学]