文献类型：期刊文献

中文题名：基于一维卷积神经网络的儿童身体活动类型识别模型构建研究

英文题名：Construction of Recognition Model for Children’s Physical Activity Types Based on One-Dimensional Convolutional Neural Network

作者：黄彩云[1];陈德武[2];何吉福[1];胡艺[1];王楠[1];陈沛[1]

第一作者：黄彩云

机构：[1]甘肃中医药大学,甘肃兰州730000;[2]中国石油勘探开发研究院西北分院,甘肃兰州730020

第一机构：甘肃中医药大学

年份：2023

卷号：59

期号：6

起止页码：10

中文期刊名：中国体育科技

外文期刊名：China Sport Science and Technology

收录：CSTPCD;;国家哲学社会科学学术期刊数据库;北大核心:【北大核心2020】；

基金：甘肃省教育科学“十四五”规划2021年度“双减”专项课题(GS〔2021〕GHBZX269)。

语种：中文

中文关键词：儿童;身体活动;动作识别;深度学习;一维卷积神经网络

外文关键词：children;physical activity;action recognition;deep learning;one-dimensional convolutional neural network

摘要：目的:通过构建高精度、高效的儿童身体活动类型识别模型,为监控儿童日常身体活动及肥胖预防等提供科学、有效的工具。方法:基于包含10种儿童身体活动类型(慢走、快走、慢跑、快跑、走上楼梯、走下楼梯、跳绳、站起、坐下、保持静止)的三轴加速度计数据集,设计了计算复杂度较低的一维卷积神经网络结构ConvNet1D-4,对数据集中10种儿童身体活动类型通过不同的组合方式开展了网络模型的训练和分类研究,并与以往研究成果进行了比较分析。结果:一维卷积神经网络模型ConvNet1D-4对10种儿童身体活动类型分类平均准确率为91.9%,合并2种加速度计数据相似的身体活动生成9种活动类型时,平均准确率为99.5%,均优于前人研究结果,且模型的计算复杂度更低。结论:基于一维卷积神经网络的儿童身体活动类型识别模型ConvNet1D-4性能优良,分类效率高,可在儿童日常身体活动的监控中实现规模化应用。
Objective:To provide a scientific and effective means for monitoring children’s daily physical activity and preventing obesity,etc.,by constructing a high-precision and efficient recognition model of children’s physical activity types.Methods:Based on a triaxial accelerometer dataset containing 10 types of children’s physical activity(slow walking,fast walking,jogging,speedy running,walking upstairs,walking down stairs,skipping rope,standing up,sitting down,and remaining stationary),ConvNet1D-4,a one-dimensional convolutional neural network structure with low computational complexity was designed,and the training and classification research of network model were conducted for the 10 types of children’s physical activity in the dataset through different combinations.The results were compared and analyzed with those of previous research.Results:The one-dimensional convolutional neural network model ConvNet1D-4 achieves an average accuracy of 91.9%in classifying 10 types of children’s physical activity.When combining two physical activities with similar accelerometer data to generate nine activity types,the average accuracy of the ConvNet1D-4 is 99.5%in classifying 10 types of children’s physical activity.Both the two results are better than those of previous research methods,and the computational complexity of the model is also lower than that of previously trained models.Conclusions:ConvNet1D-4,the recognition model of children’s physical activity types based on the one-dimensional convolutional neural network has excellent performance and high classification efficiency,which can be applied in large in monitoring children’s daily physical activities.