可穿戴生理传感器驱动的深度学习情绪识别模型在心理健康评估中的应用  

Application of Wearable Physiological Sensor-Driven Deep Learning Emotion Recognition Model in Mental Health Assessment

作　　者：石敏 许莉钧[2] 彭华东 SHI Min;XU Lijun;PENG Huadong(School of Education,Fuzhou University of International Studies and Trade,Fuzhou 350202,China;Institute of Art and Design,Nanjing Institute of Technology,Nanjing 211167,China;School of Art and Design,Fuzhou University of International Studies and Trade,Fuzhou 350202,China)

机构地区：[1]福州外语外贸学院教育学院,福州350202 [2]南京工程学院艺术与设计学院,南京211167 [3]福州外语外贸学院艺术与设计学院,福州350202

出　　处：《西南大学学报（自然科学版）》2024年第12期189-201,共13页Journal of Southwest University(Natural Science Edition)

基　　金：国家自然科学基金项目(61702225);2023年江苏省研究生教育教学改革项目(2023YJYJG02);福州外语外贸学院课题(FWKQJ201910)。

摘　　要：准确的情绪识别对心理健康问题的早期诊断和干预具有重要意义.可穿戴生理传感器在情绪识别中展现出创新性的技术应用.然而,整合多种生理信号检测情绪是一项复杂而充满挑战的任务.针对这些挑战,提出了一种深度学习模型驱动的时间多模态融合方法,以捕捉脑电图(EEG)和血容量脉搏(BVP)信号之间以及内部的非线性情绪相关性,并提高情绪分类性能.采用端到端的时间多模态深度学习模型,将来自轻量级传感器的EEG和BVP信号进行融合,以执行情绪识别任务.通过使用卷积神经网络(ConvNet)和长短期记忆网络(LSTM)模型,研究整合EEG和BVP信号,共同学习并探索跨模态高度相关的表示形式.通过智能可穿戴传感器收集数据集验证时间多模态融合方法,并与最新研究结果进行比较.实验结果表明,该方法在情绪唤醒度估计上取得了89.16%的准确率,与其他先进方法达到了相似水平.将该方法应用于焦虑治疗评估,以验证深度学习技术在心理健康应用中的有效性.实验结果表明,该方法成功提取了愉悦和唤醒估计值,有效评估了时间域内的不同情绪变化,为心理健康研究和治疗提供了有益的参考.Accurate emotion recognition has significant implications for the early diagnosis and intervention of mental health issues.Wearable physiological sensors have exhibited innovative technological applications in emotion recognition.However,integrating various physiological signals to detect emotions is a complex and challenging task.In response to these challenges,this paper proposes a deep learning model-driven time multimodal fusion approach to capture the nonlinear emotion correlations both internally and between electroencephalogram(EEG)and blood volume pulse(BVP)signals,and enhance emotion classification performance.The study employs an end-to-end time multimodal deep learning model,fusing EEG and BVP signals from lightweight sensors to perform emotion recognition tasks.Specifically,by utilizing Convolutional Neural Network(ConvNet)and Long Short-Term Memory(LSTM)models,this research integrates EEG and BVP signals,jointly learning and exploring highly correlated representations across modalities.The proposed multimodal fusion method is validated using a dataset collected through intelligent wearable sensors and compared with the latest research results.Experimental results demonstrate that the proposed method achieves an accuracy of 89.16%in estimating arousal levels,reaching a comparable level to other advanced methods.Additionally,this study applies the method for anxiety therapy assessment to validate the effectiveness of deep learning technology in mental health applications.Experimental results indicate that the proposed method successfully extracts valence and arousal estimates,and effectively evaluates different emotional changes in the time domain,providing valuable insights for mental health research and treatment.

关 键 词：情绪识别 深度学习 智能穿戴 生理信号 心理健康 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]