Study On Aerobic Exercise Fatigue Based On Bioelectrical Signals

Recently,as the development of economy in china,the Chinese people's quality of life has improved a lot.Meanwhile,sports and fitness have played more important roles in people's daily life.For all the common people,professional athletes and soldiers who need high load training,the scientific arrangement of exercise like training intensity,frequency and fatigue level monitoring to avoid excessive fatigue injury or sudden death has become an important issue.Focus on these problems,in this paper we propose to use bioelectrical signals to monitor the aerobic exercise fatigue of male by analyzing people's bioelectrical signals including ECG(Electrocardiogram),EMG(Electromyogram)and EEG(Electroencephalogram)under different physical fatigue states and distinguishing their recovery period from aerobic exercise fatigue in an objective way.The main work of this paper is as follows:1.In this paper,we designed the paradigm of aerobic exercise fatigue experiment,and collected the ECG,EMG and EEG signals from 39 male subjects.Then the fatigue scale is proposed as the reference value of fatigue degree by evaluating the subject's feeling of fatigue during exercise.2.Then,an adaptive filter was designed to preprocess the collected ECG,EMG and EEG signals as these signals contained too much artificial noise.During the experiment,the acceleration data with three-axis which can show the state of the subject's movement simultaneously,is acquired as the reference signal and is combined with ECG signal together as the input.The filter uses the adaptive characteristics to continuously optimize its parameters and weights,so that it can automatically match the change in noise and restrain the noise effectively.Additionally,adaptive differential threshold is combined with adaptive amplitude threshold to identify the R wave peaks in ECG signals and extract the heart rate variability.As for EMG and EEG signals,we mainly use bandpass filter,notch filter,wavelet filter,threshold filter and other methods to process the noise.3.Based on the changes of these characteristics under different fatigue states,discriminative features were extracted from ECG,EMG and EEG signals.The EMG characteristics were normalized to reduce individual differences.The RR interval standard deviation,low-frequency power/high-frequency power,approximate entropy,normalized rectified average value of EMG characteristics and normalized total power were selected as main features to build the model for different motor fatigue state recognition.Also,the application of multiple physiological parameters in bioelectrics to identify fatigue can avoid effect such as low recognition rate and poor applicability which is caused by only using single physiological index to study fatigue in some studies.4.A motion fatigue classifier based on support vector machine algorithm and random forest algorithm is designed by using the features extracted above.Then their classification accuracy is tested by 5-fold cross-validation method.The support vector machine fatigue classifier algorithm is traditional,reliable,simple and efficient.The random forest fatigue classifier algorithm is based on a decision tree.If the random forest produce wrong classification results,it requires more than half of the decision trees to make the wrong decision and at the same time the number of votes for a wrong classification is higher than that of the correct classification,so it has a low classification error rate and strong robustness.After 5-fold cross-validation,the recognition rate of fatigue degree of SVM and random forest classifier reached 91.39% and 95.15% respectively.5.Finally,the changes of ECG and EEG during fatigue recovery period were analyzed,and the model for fatigue recovery identifying based on SVM was designed by using the ECG signal of the last 2 minutes of rest period after exercise to detect the rest recovery of participants.