Non-invasive Physical Fatigue Detection Based On Deep Learning

As an important physiological indicator of the body,physical fatigue has attracted more and more attention.In the past few decades,the effects of physical fatigue on human health and specific diseases have been extensively studied.Due to the different respond to physical fatigue of different people,physical fatigue testing is becoming more and more important in the field of physical health.This paper focuses on non-invasive physical fatigue testing based on deep learning and achieving non-invasive,intelligent,high-precision physical fatigue testing.By analyzing the extracted feature results of the face region ROI to select the optimal ROI and then the multi-spectral imaging technique is used to extract the muscle motion signal characteristics of the facial ROI,and the LSTM network model is constructed according to the sequence change of the facial muscle motion signal characteristics.We trained this model to classify the physical fatigue state and baseline state.At the same time,it is compared with the traditional classification algorithm in terms of physical fatigue detection as a reference.The main contributions of this paper in this work are as follows:(1)A video-like hyperspectral data set for physical fatigue detection was obtained by recruiting subjects participating in the experiment.(2)The FMAD algorithm is proposed to extract real-time muscle fatigue signal features from the face of the subjects in the dataset using hyperspectral imaging techniques.(3)Through a large number of experiments to analyze the sensitivity of ROI to fatigue signal characteristics,select the ROI with the greatest sensitivity to fatigue characteristic signals,and prepare for the establishment of physical fatigue detection model.(4)Extracting classification features by changes in muscle trajectory movement in the ROI.These features are mapped into the frequency domain space to obtain the fatigue characteristics of the body.(5)Propose the LSTM-based physical fatigue detection algorithm,first use TensorFlow to build an LSTM network model,and then use the previous physical fatigue feature sequence to train the model in the future until the model converges and the classification test through the test set is very good.Effect.The LSTM-based non-contact physical fatigue detection identification is finally achieved,and the participant’s background data or baseline status information is not required.The last proposed LSTM based physical fatigue detection algorithm has achieved good results in experiments,with an accuracy rate of 75%,which laid the foundation for future industrialization.Experimental results show that multispectral imaging,as a non-invasive method,has the potential to recognize physical fatigue.