基于多域特征参数的长期屈颈位人群颈部疲劳识别分析与研究

Analysis and Research on Neck Fatigue Identification in Long-term Flexed Neck Position Population Based on Multi-domain Characteristic Parameters

长时间保持屈颈姿态易导致颈肌疲劳,从而引起颈椎的软组织损伤和退行性改变。因此,对长期屈颈伏案工作者的颈部肌肉疲劳状态进行识别具有很高的研究价值。首先,采用小波阈值法、经验模态分解法(empirical mode decomposition,EMD)、EMD与小波阈值结合的方法对长时间屈颈伏案人员的颈部肌电信号进行消噪处理,并通过计算均方根误差、信噪比及运算速度对3种去噪方法进行对比分析,得出基于EMD与小波阈值的消噪方法的信噪分离能力最好。其次,提取了颈部肌电信号的时域、频域及非线性特征参数,通过分析各特征参数随时间变化趋势确定有效特征。最后,将有效特征向量作为分类器输入,建立卷积神经网络(convolutional neural network,CNN)模型对颈部肌肉的疲劳状态进行识别,分类准确率达到90.48%。

Maintaining the flexion position for a long time can easily lead to neck muscle fatigue,resulting a soft tissue injury and degenerative changes of the cervical spine.Therefore,it is significant to discriminate the fatigue state of the neck muscles of the workers who bend and sit at the desk for a long time.The wavelet threshold method,Empirical Mode Decomposition(EMD),the combination of EMD and wavelet threshold are used to de-noise the neck EMG of the people who bend the neck for a long time.By calculating root mean square error(RMSE),signal-to-noise ratio(SNR)and operation speed,the three denoising methods are compared and analyzed.The results show that the denoising method the combination of EMD and wavelet threshold has the best signal-to-noise separation ability.Then the time domain,frequency domain and nonlinear characteristic parameters of the neck EMG signal are extracted,and the effective characteristics are determined by analyzing the variation trend of each characteristic parameter with time.Finally,the effective feature vector is used as the input of the classifier,and the convolutional neural network model is established to recognize the fatigue state of the neck muscle,and the classification accuracy reaches 90.48%.