Research On Fatigue State Detection Method Based On Multi-feature Fusion

The fatigue state usually manifests as slow response,inattention,and easy distraction.Long-term fatigue will not only interfere with daily work efficiency,but also easily cause safety accidents.Therefore,if the fatigue of personnel can be effectively detected,the safety of personnel can be improved.Based on computer vision technology,this thesis constructs a set of fatigue detection methods based on face recognition.Classification of fatigue expressions based on deep learning models,and use of information fusion technology to build fatigue judgment models.The research innovations of the thesis are as follows:Considering that the deep learning model requires a large amount of experimental data,in many cases,the amount of face image data cannot meet the experimental needs.Therefore,based on the Inception-Resnet v1 network model,this thesis designs a small sample face comparison method based on the twin neural network and the comparison loss function,and realizes face recognition.The experimental accuracy is the performance indicator,and it is compared with the other 4 kinds of face recognition.Model comparison shows that the algorithm proposed in this thesis achieves an accuracy of 98.12% for face recognition in a small sample set.The research is based on deep learning technology to realize the detection of fatigue expression.On the personal fatigue expression detection data set,based on Inception-Resnet v1,the network model is modified and deleted so that the model achieves better results than the traditional Resnet50,Inception V4,VGG16 and Inception-Resnet v2 models;and in order to improve The accuracy of the model,by comparing the existing attention mechanism structure,chooses the module based on the hybrid domain attention mechanism CBAM as the prototype to embed it on the top of the network,so that the network model embedded with the attention mechanism has better performance;The accuracy is a performance indicator.Compared with the other four fatigue detection algorithms,the algorithm proposed in the article achieves an accuracy of 97.82% for the fatigue expression detection of a single frame image.Considering that a single index cannot well consider all aspects of fatigue determination,and it cannot quantify the degree of fatigue.Therefore,based on the information fusion technology based on the analytic hierarchy process,this thesis establishes a fuzzy fatigue evaluation rule,expressing the fatigue degree in percentage;using the SVM classification model test to verify the effectiveness of the analytic hierarchy process.