Research On Intelligent Recognition Of Human Fatigue State And System Lightweight

With the development of society,more and more people are tired from work,irregular daily life,and lack of exercise to keep the body in a sub-health state for a long time.Therefore,proper exercise is very important.Real-time understanding of your own muscle fatigue during exercise and timely adjustment of exercise intensity can effectively improve the scientific nature of exercise.This paper designs and produces a set of intelligent recognition system for human fatigue status.The surface EMG signal data of the six muscles of the legs during running were collected by the s EMG acquisition device,sent to the PC through the master-slave Bluetooth module HC-05,and loaded on the PC.The trained fatigue discrimination model performs fatigue discrimination on the collected s EMG data,and synchronizes the fatigue state to the cloud database in real time,makes a mobile terminal APP,updates the fatigue state in real time,and analyzes the fatigue state statistics during the exercise period,and Provide users with reasonable exercise suggestions to improve the scientific nature of user exercise.Collect the s EMG data of the six muscles of the legs in the middle-and-long-distance running.800-1200,1200-1500 m running distance for fatigue labeling,using this as input to build a CNNbased deep learning model and machine learning algorithm model to perform fatigue state classification training on collected EMG signals,Through experimental comparison,CNN-RF model has the best effect in analyzing fatigue state.The model can perform high-precision fatigue state recognition on the test set,and the correct recognition rate reaches 92%.Compared with other commonly used classification models,the accuracy rate of fatigue discrimination in this scenario is higher.Dimension reduction of the data set can greatly reduce the model parameters and FLOPs while reducing the noise in the data set.The model parameters are reduced from1038.96 M to 260.41 M,and the FLOPs are changed from 5.192 G to 1.299 G,and the fatigue judgment is accurate.The rate increased to 96%.Structured pruning of the CNN-based feature extraction model can further compress the overall model under the premise that it has little impact on the accuracy of recognition to achieve lightweighting of the model.Production of mobile phone terminal APP.The APP mainly includes a registration and login module,a status display module,and a fatigue analysis module.The generalized universal ability of the overall fatigue discrimination system was tested experimentally.None of the selected test personnel participated in the model training data collection experiment.The implementation results show that the system model based on CNN-RF has the highest recognition accuracy,which is 76.75%.After analysis,it is concluded that the intelligent recognition system for human exercise fatigue state can determine the fatigue state during running in real time.However,the current system model has a poor generalization ability,and the accuracy of the system for 0 state and 3 state fatigue discrimination is higher.