基于人机交互的心理健康监测数据异常标记识别研究  

Study on abnormal marker identification of mental health monitoring data based on human-computer interaction

作　　者：任倩[1] 王博[1] REN Qian;WANG Bo(College of economics and Management,Shangluo University,Shangluo,Shanxi 726000,China)

机构地区：[1]商洛学院,陕西商洛726000

出　　处：《自动化与仪器仪表》2023年第7期182-186,共5页Automation & Instrumentation

基　　金：陕西高校网络思想政治工作研究课题与实践项目《“爱心理幸福空间站”网络心理健康教育实践》(2022WSYJ100073);商洛学院2022年学生工作研究课题《“爱心理幸福空间站”网络心理健康教育研究》(XSGZ2207);陕西普通高校辅导员工作室建设成果《“向阳花”大学生危机预控工作室》(2019GZS23)。

摘　　要：针对人机交互的心理健康异常识别问题,提出以人脸表情监测和分类作为基础参数,建立一个基于多层特征融合的多类别表情识别模型。首先,对原始面部表情图像进行姿态和灰度归一化;然后以VGGNet网络对模型进行轻量化,将网络浅层局部特征与深层全局特征进行融合;之后利用SoftMax分类器进行表情识别和分类;最终基于表情识别结果实现心理健康数据异常监测。结果表明,采用VGGNet网络与模块3和模块4进行多层融合可取得78.2%的准确率,基于此融合方式可实现表情识别模型搭建。对数据集进行平衡和增广处理后,模型的表情识别准确率进而由78.2%提升至85.3%。本算法的表情识别准确率可高达99.67%,比传统的SVM分类算法和改进AlexNet卷积神经网络分别高出了9.62%和8.05%。且本算法可对9种不同类型的表情进行实时分类,为心理健康监测系统提供了有效数据支撑。Aiming at the recognition of mental health abnormalities in human-computer interaction,a multi-category expression recognition model with facial expression monitoring and classification based on multi-layer feature fusion.First,posture and gray normalization;then the model is lightweight by the VGGNet network to integrate shallow local features with deep global features;then use the SoftMax classifier for expression recognition and classification;and finally realize abnormal monitoring of mental health data based on the expression recognition results.The results show that the VGGNet network,module 3 and module 4 can achieve 78.2%accuracy,and the expression recognition model can be built based on this fusion method.After balancing and enlarging the data set,the expression recognition accuracy of the model was further increased from 78.2%to 85.3%.The expression recognition accuracy of this algorithm can be as high as 99.67%,which is 9.62%and 8.05%higher than the traditional SVM classification algorithm and the improved AlexNet convolutional neural network,respectively.Moreover,this algorithm can classify 9 different types of facial expressions in real time,providing effective data support for the mental health monitoring system.

关 键 词：人机交互 心理健康监测 表情识别 VGGNet网络 多层特征融合 

分 类 号：TP392[自动化与计算机技术—计算机应用技术]