【作者】 王恒；

【导师】 柯丽；

【作者基本信息】 沈阳工业大学 ， 工程硕士（专业学位）， 2022， 硕士

【摘要】 肌肉是重要的生物组织,支持着人体各部分的运动。当机体承受较大负荷时,其结构会发生相应变化,从而表现出不同程度的机能状态。在日常生活中,肌肉疲劳已经成为一种常见的疾病,引起了康复医学、运动力学和人体工效学专家学者的重视。肌肉疲劳是肌肉组织的重要运动特征,通常情况下,肌肉疲劳是可逆的,如果不及时治疗,可能会导致永久性损伤,严重影响人们的工作、锻炼和日常生活。因此,本文以生物电阻抗测量技术为基础,提出了一种以肌肉细胞内外液生物阻抗及容量变化来表征肌肉疲劳程度的方法,并利用基于机器学习的集成算法对运动肌肉状态进行了预测分类。首先依据细胞内外液理论及肌肉疲劳生理基础,分别采集在高频激励下和低频激励下的肌肉疲劳过程生物阻抗信号,利用一维高斯滤波对所有信号进行预处理,提取平均阻抗值,完成数据的收集工作。并依据预处理完后的时序信号特征,给出细胞内外液阻抗及容量与肌肉疲劳的定性关系。其次,采用人体分段阻抗测量模型,给出手臂肌肉的细胞内液与细胞外液阻抗计算公式。对高斯去噪后高频激励及低频激励下的肌肉阻抗信号进行特征值提取,并对肌肉疲劳前后细胞内外液的平均阻抗值进行了计算与分析。利用改进Moissl方程对肌肉疲劳前后细胞内外液容量进行了计算与定性分析。实验结果显示,肌肉疲劳后相较于静息状态,细胞外液的,阻抗值下降,容量增加;细胞内液的阻抗值升高,容量减少。细胞内外液之和即总体液略有下降。最后,提出了基于机器学习的肌肉疲劳状态分类集成算法,选取了K近邻（KNN,K-Nearest Neighbor）、决策树（DT,Decision Tree）、支持向量机（SVM,Support Vector Machine）、神经网络（DNN,Deep Neural Networks）四个基分类器。根据实验中所测得的阻抗信号进行特征值提取,建立双频融合的数据集,将肌肉在静态运动性等长收缩下分类为三种状态,即S₁非疲劳态（静息状态）、S₂疲劳过渡态（中度疲劳）、S₃深度疲劳态（极其疲劳但非力竭状态）。并利用五折交叉验证进行训练提高模型的泛化能力。通过静态运动性疲劳实验及其数据分析,结果显示,本文算法具有较高的分类准确率,从而证明了该分类模型可以在局部疲劳中对肌肉运动状态进行预测。更多还原

【Abstract】 Muscle is an important biological tissue that supports the movement of all parts of the human body.When the body bears a large load,its structure will change accordingly,thus showing different degrees of functional state.In daily life,muscle fatigue has become a common disease,which has attracted the attention of experts and scholars in rehabilitation medicine,sports mechanics and ergonomics.Muscle fatigue is an important sports feature of muscle tissue.Usually,muscle fatigue is reversible.If it is not treated in time,it may lead to permanent injury and seriously affect people’s work,exercise and daily life.Therefore,based on bioelectrical impedance measurement technology,this paper proposes a method to characterize the degree of muscle fatigue by the changes of biological impedance and capacity of fluid inside and outside muscle cells,and uses the integrated algorithm based on machine learning to predict and classify the state of moving muscle.Firstly,according to the theory of intracellular and extracellular fluid and the physiological basis of muscle fatigue,the biological impedance signals of muscle fatigue process under high-frequency excitation and low-frequency excitation are collected respectively.All signals are preprocessed by one-dimensional Gaussian filter to extract the average impedance value and complete the data collection.According to the time sequence signal characteristics after preprocessing,the qualitative relationship between intracellular and extracellular fluid impedance and capacity and muscle fatigue is given.Secondly,using the human body subsection impedance measurement model,the calculation formulas of intracellular and extracellular fluid impedance of arm muscle are given.The characteristic values of muscle impedance signals under high-frequency excitation and low-frequency excitation after Gaussian denoising were extracted,and the average impedance values of intracellular and extracellular fluid before and after muscle fatigue were calculated and analyzed.The volume of intracellular and extracellular fluid before and after muscle fatigue was calculated and qualitatively analyzed by using the improved moissl equation.The experimental results showed that after muscle fatigue,compared with the resting state,the impedance value of extracellular fluid decreased and the capacity increased;The impedance value of intracellular fluid increased and the capacity decreased.The sum of intracellular and extracellular fluid,that is,the total fluid,decreased slightly.Finally,an integrated algorithm of muscle fatigue state classification based on machine learning is proposed,and four base classifiers are selected:KNN,DT,SVM and DNN.According to the impedance signal measured in the experiment,the eigenvalues are extracted,and the dual frequency fusion data set is established.The muscle is classified into three states under static motor isometric contraction,namely S₁non fatigue state（resting state）,S₂fatigue transition state（moderate fatigue）and S₃deep fatigue state（extremely fatigue but not exhausted state）.The model is trained by 50%cross validation to improve the generalization ability of the model.Through static exercise fatigue experiments and data analysis,the results show that the algorithm in this paper has a high classification accuracy,which proves that the classification model can predict the muscle movement state in local fatigue.更多还原