Methodological Study of Visual Pulse Diagnosis Based on Multi-Level Task
Pulse diagnosis is crucial for Chinese medicine diagnosis.However,the existing objectivization of pulse diagnosis highly relies on equipment,which,unfortunately,hinders its promotion in mobile health.Inspired by the recent video-based non-contact physiological measurements based on remote photoplethysmography(rPPG),this paper proposes an rPPG-based"visual pulse diagno-sis"method to predict the pulse with facial video.The study performed ROI detection through multi-scale fusion and multi-task learn-ing,obtained rPPG-wave by filtering based on deep convolutional neural networks,and conducted pulse prediction using deep neural networks.After examination in real cases,we found that the accuracy of the deep learning and multilevel task-based"visual pulse di-agnosis"can be over 85%and outperforms other rPPG methods."Visual Pulse Diagnosis"is an innovative paradigm for non-contact pulse diagnosis,which can be applied to mobile healthcare in Chinese medicine in the future with promising value.
Visual pulse diagnosisMultilevel taskObjectivization of pulse diagnosisrPPG
万方数据
