电气技术  2018, Vol. 19 Issue (11): 26-30    DOI:
研究与开发 |
LSTM递归神经网络人体活动行为识别算法研究
李智敏, 刘一鹏, 郑海峰, 冯心欣
福州大学物理与信息工程学院,福州 350116
Research on an LSTM recurrent neural network algorithm for human activity recognition
Li Zhimin, Liu Yipeng, Zheng Haifeng, Feng Xinxin
College of Physics and Information Engineering, Fuzhou University, Fuzhou 350116
全文: PDF (15779 KB)   HTML (1 KB) 
输出: BibTeX | EndNote (RIS)      
摘要 为克服支持向量机(SVM)等机器学习算法在人工提取数据特征时特征提取不充分和信息丢失的问题,本文提出基于长短期记忆(LSTM)的递归神经网络算法实现对人体活动行为模式的识别。本文提出的算法不仅能充分提取所收集到的人体活动行为数据的时间特性,还可有效避免传统递归神经网络的梯度消失问题,从而提高了人体活动行为识别的准确性。本文使用WISDM数据集对神经网络进行训练和测试,并采用交叉验证的方法对所提出的模型进行评估。实验结果表明,本文提出的方法在识别人体活动行为时相比较于SVM等机器学习算法提升了识别的精确度。
服务
把本文推荐给朋友
加入我的书架
加入引用管理器
E-mail Alert
RSS
作者相关文章
李智敏
刘一鹏
郑海峰
冯心欣
关键词 递归神经网络LSTM活动行为识别移动感知    
Abstract:In order to overcome the problem of insufficient feature extraction and lost information incurred by the traditional machine learning based methods such as support vector machine (SVM), we propose a long short-term memory (LSTM) based neural network algorithm to recognize human activity in this paper. The proposed algorithm can not only fully extract time characteristics of human activity data but also effectively avoid the problem of gradient disappearance in traditional recurrent neural network, thus improving the accuracy of human activity recognition. In this paper, the proposed approach is trained and tested by using a WISDM dataset and ievaluated by the cross validation method. Experimental results show that the proposed method can improve the accuracy of recognition when comparing with the other machine learning algorithms such as SVM.
Key wordsrecurrent neural network    LSTM    activity recognition    mobile sensing   
收稿日期: 2018-06-13      出版日期: 2018-11-16
作者简介: 李智敏(1994-),男,福建省莆田市人,硕士研究生,研究方向为群智感知网络中的数据分析。
引用本文:   
李智敏, 刘一鹏, 郑海峰, 冯心欣. LSTM递归神经网络人体活动行为识别算法研究[J]. 电气技术, 2018, 19(11): 26-30. Li Zhimin, Liu Yipeng, Zheng Haifeng, Feng Xinxin. Research on an LSTM recurrent neural network algorithm for human activity recognition. Electrical Engineering, 2018, 19(11): 26-30.
链接本文:  
http://dqjs.cesmedia.cn/CN/Y2018/V19/I11/26