基于百度Paddle的人类活动识别研究  

Research on human activity identification based on Paddle

作　　者：赖志伟 邓述为 Lai Zhiwei;Deng Shuwei(College of Information Engineering,Hunan University of Science and Engineering,Yongzhou 425199,China)

机构地区：[1]湖南科技学院信息工程学院,湖南永州425199

出　　处：《无线互联科技》2023年第2期141-143,共3页Wireless Internet Technology

摘　　要：文章详细阐述了一个基于机器学习的人类活动识别方法。该方法对人类活动数据进行探索和预处理,提出了一个对人类活动进行识别的LSTM模型。本文中的问题是一个典型的分类问题,目标变量是6种不同种类的人类活动,选择准确率作为模型的评测指标。具体方法是,通过读取训练阶段保存到本地的模型,以相同的数据构造方式对测试集进行预测评估,不断地调整学习率参数。研究表明,本文提出的LSTM模型在迭代10次、隐层数为50、学习率为0.01时达到了比较好的准确率,在测试集上有比较好的表现。This paper describes a machine learning based human activity recognition method in detail.This method explores and preprocesses human activity data,and proposes a LSTM model for human activity recognition.The problem in this paper is a typical classification problem.The target variables are six different kinds of human activities,and the accuracy rate is selected as the evaluation index of the model.Specific approach is to save to a local model by reading the training phase,and then by the same data structure was carried out on the test set predictive evaluation,then vector by constantly adjusting parameters.The research indicates that the proposed LSTM model is in the iteration vector 10 times,the number of hidden layer of 50 and achieves a better accuracy of 0.01,having a good performance on the test set.

关 键 词：机器学习 人类活动识别 统计分析 异常预警 

分 类 号：TN37[电子电信—物理电子学]