摘要
针对健康数据种类日益增多,而统计学算法不能实现所有数据种类的特征提取与健康状态评估的问题,文中提出了基于卷积与BP神经网络的健康数据分析算法来评估用户的健康状态。对健康数据类型进行分析,总结为数字、文本、图像3种模态的数据类型,并分别针对这3种数据类型进行基于卷积神经网络的数据特征表征模型的构建。经过特征融合,利用多元高斯分布定义健康状态的划分,并利用BP神经网络构建健康数据分析算法。通过在样本数据上的测试结果表明,与朴素贝利斯模型对比,文中所述健康数据分析算法具有较高的准确率,使用多模态数据较单一数据类型的健康评估结果更优,其准确率约为84.2%。
In view of the increasing variety of health data,but the statistical algorithm can not achieve the problem of feature extraction and health status evaluation of all data types,this paper proposes a health data analysis algorithm based on convolution and BP neural network to evaluate the health status of users.Health data types are analyzed,which are summarized into three modes:number,text and image.The data feature representation model based on convolutional neural network is constructed for these three types of data.After feature fusion,the multi Gaussian distribution is used to define the division of health state,and BP neural network is used to build the health data analysis algorithm.The test results on the sample data show that,compared with the naive Bayliss model,the health data analysis algorithm described in this paper has a higher accuracy,and the results of health assessment using multimodal data are better than that of single data type,about 84.2%.
作者
李化奇
张静
魏涛
LI Huaqi;ZHANG Jing;WEI Tao(Jinan City People’s Hospital,Laiwu 271100,China)
出处
《电子设计工程》
2020年第23期38-42,共5页
Electronic Design Engineering
基金
2018教育部高教司协同育人项目(201802362013)。
关键词
健康数据分析算法
卷积神经网络
BP神经网络
多模态数据
特征融合
特征表征
health data analysis algorithm
convolutional neural network
BP neural network
multimodal data
feature fusion
feature representation
