基于贝叶斯网络建模的疼痛评估研究  

Study on pain assessment based on Bayesian network

作　　者：郭文强[1] 赵艳[2] 张栋 黄梓轩 侯勇严[2] 肖秦琨[3] 郭志高 GUO Wen-qiang;ZHAO Yan;ZHANG Dong;HUANG Zi-xuan;HOU Yong-yan;XIAO Qin-kun;GUO Zhi-gao(School of Electronic Information and Artificial Intelligence,Shaanxi University of Science&Technology,Xi′an 710021,China;School of Electrical and Control Engineering,Shaanxi University of Science&Technology,Xi′an 710021,China;School of Electronic Information Engineering,Xi′an Technological University,Xi′an 710021,China)

机构地区：[1]陕西科技大学电子信息与人工智能学院,陕西西安710021 [2]陕西科技大学电气与控制工程学院,陕西西安710021 [3]西安工业大学电子信息工程学院,陕西西安710021

出　　处：《陕西科技大学学报》2021年第6期161-166,共6页Journal of Shaanxi University of Science & Technology

基　　金：国家自然科学基金项目(62071366);陕西省科技厅重点研发计划项目(2020SF-286);陕西省教育厅产业化计划项目(18JC003);陕西省西安市科技计划项目(2019216514GXRC001CG002GXYD1.1)。

摘　　要：复杂环境下的不确定性因素给面部表情疼痛评估的分析和建模带来了巨大的挑战.针对疼痛表情识别中的复杂、不确定性问题,提出了一种基于贝叶斯网络(BN)建模的疼痛评估方法.首先对人脸图像获取有关疼痛的面部动作单元(AU).其次在分析了疼痛与面部AU关系的基础上提出了一种疼痛评估BN结构.采用特征样本训练得到BN模型参数,建立疼痛评估BN模型.最后利用BN推理算法实现疼痛评估.实验结果表明:在完全证据条件下,与经典的支持向量机、多示例学习、循环神经网络方法相比,该方法有着更高的识别率;在不完全证据条件下,即使有关疼痛的AU特征向量存在局部缺失,本文方法仍具有较好的疼痛评估结果,并为不确定性环境下的疼痛识别提供了一种有效的途径.Uncertainty factors in complex environments have brought huge challenges to the analysis and modeling of facial expression pain assessment.Aiming at the problem of pain expression recognition in a complex and uncertain environment,a pain assessment method based on Bayesian Network(BN)modeling is proposed.Firstly,the facial Action Unit(AU)related to pain is obtained from the face image.Secondly,based on the analysis of the relationship among pain and facial AU,a BN structure for pain assessment is proposed.The BN model parameters are obtained by feature sample training,and the pain assessment BN model is established.Finally,the BN reasoning algorithm is used to achieve pain assessment.The experimental results show that:Under the condition of complete evidence,compared with the classic Support Vector Machine,Multi-Instance Learning,and Recurrent Neural Network methods,this method has a higher recognition rate;under the condition of incomplete evidence,even if the pain-related AU feature vector is partially missing,the method in this paper still has a good pain assessment result,and provides an effective way for pain recognition in an uncertain environment.

关 键 词：疼痛评估 贝叶斯网络建模 BN推理 AU特征向量 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]