基于深度学习的高空坠落危险行为识别方法
A Method for Identifying High-altitude Falling Hazard Behavior Based on Deep Learning
聂程 1叶翔 1方百里 1孙嘉兴 1张滔1
作者信息
- 1. 广东电网有限责任公司广州供电局,广东 广州 510180
- 折叠
摘要
以卷积神经网络为代表的深度学习算法可以更加精准有效地提取人体行为特征,因此将深度学习用于人体行为识别与预测成为研究热点.文章在经典HRnet网络结构的基础上通过改进L-Swish激活函数和引入Squeeze-and-Excitation模块,提出一种新型网络模型L-HRnet,用于判断施工人员高空作业时的行为动作是否存在危险性.在公开数据集HMDB51 上进行行为分类与识别实验,结果表明,改进后网络结构L-HRnet的识别准确率明显优于HRnet,有效提升了高空作业人员的防护水平.
Abstract
Deep Learning algorithms represented by Convolutional Neural Networks can extract human behavior features more accurately and effectively,applying Deep Learning to human behavior recognition and prediction has become a research hotspot.On the basis of the classic HRnet network structure,this paper proposes a new network model L-HRnet by improving the L-Swish activation function and introducing the Squeeze-and-Excitation module,which is used to determine whether the behavioral actions of construction worker during high-altitude operations are dangerous.Behavioral classification and recognition experiments are conducted on the public dataset HMDB51,and the results show that the improved network structure L-HRnet had significantly better recognition accuracy than HRnet,effectively improving the protection level of high-altitude workers.
关键词
神经网络/深度学习/高空坠落/动作识别Key words
neural network/Deep Learning/high-altitude falling/action recognition引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据