摘要
疼痛的准确评估能够为患者制定有效镇痛方案提供依据,相较于通过面部表情与接触式测量方式进行疼痛识别与评估,基于远程光体积变化描记术的疼痛识别具有非接触和客观性等优点。本研究首先从面部视频中提取由心脏跳动引起的面部生理信号;接下来,采用小波变换、窄带通滤波方法对原始信号进行去噪处理;之后,基于得到的rPPG信号计算心率变异性特征,再利用支持向量机、决策树、K-近邻以及随机森林四种机器学习方法对疼痛进行分类;在此基础上,为克服手工提取特征的弊端,采用ResNet自动捕获不同时间步长内的特征,达到疼痛评估的目的;最后,在Biovid数据库上进行模型性能验证。实验结果表明,利用RseNet实现疼痛评估效果最好,二分类准确率为66.3%,验证了基于rPPG信号的非接触式疼痛评估是有效的,为进一步研究更精确的疼痛评估算法奠定了基础。
Accurate assessment of pain can provide a basis for patients to develop an effective analgesic program.Compared with pain recognition and assessment based on facial expression and contact measurement,pain recognition based on remote optical volume change otrography has the advantages of non-contact and objectivity.In this study,the facial physiological signals were extracted from the face video,the wavelet transform and narrow-band pass filtering methods were used to denoise the original signal,the heart rate variation characteristics and heart rate gain based on pulse wave signal,and four machine learning methods of support vector machine,decision tree,K-nearest neighbor and random forest,to overcome the disadvantages of manual feature extraction,ResNet automatically captures the features in different time steps to achieve the purpose of pain assessment.Finally,the model performance was verified on the Biovid database,and the experimental results showed that the best pain assessment by RseNet,with the second classification accuracy of 66.3%,which verified that the contactless pain assessment based on rPPG signal is effective and laid the foundation for further research on more accurate pain assessment algorithm.
作者
刘丹
郝颢
李文通
刘虔诚
蔡志丹
LIU Dan;HAO Hao;LI Wentong;LIU Qiancheng;CAI Zhidan(School of Mathematics and Statistics,Changchun University of Science and Technology,Changchun 130022;School of Life Science and Technology,Changchun University of Science and Technology,Changchun 130022)
出处
《长春理工大学学报(自然科学版)》
2025年第2期134-142,共9页
Journal of Changchun University of Science and Technology(Natural Science Edition)
基金
吉林省科技发展计划项目(20240101339JC)。
