摘要
针对人体活动识别问题与其在实际情况中的应用,综合考量卷积神经网络与作为循环神经网络变体的门控循环单元,设计能自动提取传感器数据特征和记忆时序性活动数据的CNN-GRU混合神经网络模型,并予以改良。使用该模型在公开的数据集上进行实验,较其他的模型效果更加理想。在人体活动识别的处理中,CNN-GRU模型能达到预期的高准确率。在数据集时序性依赖较强的情况下,CNN-GRU模型能拥有更好的准确度和稳定性。
Aiming at the problem of human activity recognition and its utilization in practical situations,considering the CNN and the GRU,which is a variant of the recurrent neural network,the CNN-GRU hybrid neural network model that automatically extracts features of sensor data and memorizes temporal activity data is presented and improved.Using this model to conduct experiments on public dataset,the result is more ideal than other models.In the processing of human activity recognition,the CNN-GRU model can achieve the expected high accuracy.The CNN-GRU model has better accuracy and stability when the dataset is of strong temporal dependence.
作者
吴海涛
陆志平
胡晨骏
Wu Haitao;Lu Zhiping;Hu Chenjun(College of Artificial Intelligence and Information Technology,Nanjing University of Chinese Medicine,Nanjing 210046,Jiangsu,China)
出处
《计算机应用与软件》
北大核心
2021年第8期187-193,219,共8页
Computer Applications and Software
基金
教育部产学合作协同育人项目“中医药高等院校人工智能实践教育示范中心”(201801075023)
国家级大学生创新创业训练项目“健康云智囊”(201910315002X)。
关键词
卷积神经网络
门控循环单元
混合神经网络
人体活动识别
Convolutional neural network
Gated recurrent unit
Hybrid neural network
Human activity recognition
