摘要
目前,基于信道状态信息(Channel State Information,CSI)的室内摔倒检测(Fall Detection,FD)系统已被证明拥有巨大潜力,但是,不同室内布局带来的多径效应的差异往往使其无法实现跨场景使用。因此,该文提出了DA-Fall(Domain-adaptive Fall),通过结合两种自适应策略的域自适应方法来改进未标记噪声信号的泛化,从而提高对目标域的检测精度。在提出的摔倒检测系统中,引入了域鉴别器和域混淆自适应层来进行对抗性训练。首先,该算法通过引入依赖于相对值的相对鉴别器来优化对抗训练,从而更好地反映域间差异。其次,将基于多核架构的最大均值差异(Multiple Kernel Maximum Mean Difference,MK-MMD)作为域对抗损失的正则化项,进一步减小域间的边缘分布距离。实验分析表明,DA-Fall取得了比WiFall,RT-Fall,SignGAN更好的效果,在原场景与新场景中分别达到了96.83%和91.03%的检测精度。
Indoor Fall Detection(FD)systems based on Channel State Information(CSI)have been proved to have great potential,but the difference in multipath effects caused by different indoor layouts often makes it impossible to achieve cross-scene use.Therefore,we propose DA-Fall(Domain-Adaptive Fall),which combines the domain adaptive methods of two adaptive strategies to improve the generalization of unlabeled noise signals,thereby improving the detection accuracy of the target domain.In the proposed fall detection system,a domain discriminator and a domain confusion adaptive layer are introduced for adversarial training.Firstly,such algorithm optimizes adversarial training by introducing a relative discriminator that depends on relative values,so as to better reflect the differences between domains.Secondly,the Multiple Kernel Maximum Mean Difference(MK-MMD)based on multi-core architecture is used as the regularization term of domain adversarial loss to further reduce the edge distribution distance between domains.Experiments show that DA-Fall achieves better results than Wi-Fall,RT-Fall,Sign-GAN and other systems.The detection accuracy of 96.83%and 91.03%was achieved in the original scene and the new scene,respectively.
作者
马永连
张登银
MA Yong-lian;ZHANG Deng-yin(School of Internet of Things,Nanjing University of Posts and Telecommunications,Nanjing 210023,China)
出处
《计算机技术与发展》
2023年第10期86-92,共7页
Computer Technology and Development
基金
国家自然科学基金项目(61872423)
江苏省高等学校自然科学研究重大项目(19KJA180006)。
