摘要
对用户的行走、上楼、下楼、静坐、站立、躺下6种行为状态产生的陀螺仪传感器数据进行研究。通过分析局部时间段内用户的状态信息,扩充样本数据维度,将卷积神经网络模型和长短期记忆网络模型相结合,构建特征提取器,使用线性支持向量机完成分类工作。行为识别精度达到99.4%以上,每一种的行为状态识别精度均超过98%。相比于多层感知机、卷积神经网络以及长短期记忆网络,平均识别精度提升了1%-2%。相比传统的机器学习算法,例如贝叶斯、支持向量机、决策树等,平均识别精度提升了3%-4%。
The gyroscope sensor data produced by six behavior states of users,walking,upstairing,downstairing,sitting,standing and lying down,were studied.By analyzing the state information of users in local time interval,expanding the dimension of sample data,combining convolution neural network model with long short-term memory network model,a feature extractor was constructed,and the classification was completed using linear support vector machine.The accuracy of behavior recognition is over 99.4%,and the accuracy of each action state recognition is over 98%.Compared with multi-layer perceptron,convolution neural network and long short-term memory network,the average recognition accuracy is improved by 1%-2%.Compared with traditional machine learning algorithms,such as Bayesian,support vector machine and decision tree,the average recognition accuracy is improved by 3%-4%.
作者
余万里
韦玉梅
李鲁群
YU Wan-li;WEI Yu-mei;LI Lu-qun(College of Information,Mechanical and Electrical Engineering,Shanghai Normal University,Shanghai 201400,China)
出处
《计算机工程与设计》
北大核心
2019年第10期3030-3036,共7页
Computer Engineering and Design
基金
2018年度大学生创新创业训练计划基金项目(201810270167)
关键词
卷积神经网络
长短期记忆网络
人体行为识别
深度学习
模型融合
convolution neural network
long short-term memory network
human activity recognition
deep learning
model fusion
