基于块稀疏编码的物联网无源定位机器学习  

Machine learning for device-free localization in internet of things based on block sparse coding

作　　者：江河[1] 李晓茹 孙敏[2] JIANG He;LI Xiao-ru;SUN Min(Department of Computer Science and Technology,Taiyuan University,Taiyuan 030012,China;School of Computer and Information Technology,Shanxi University,Taiyuan 030006,China)

机构地区：[1]太原学院计算机科学与技术系,山西太原030012 [2]山西大学计算机与信息技术学院,山西太原030006

出　　处：《计算机工程与设计》2022年第9期2502-2510,共9页Computer Engineering and Design

基　　金：国家自然科学基金项目(61872226);山西省高等学校教学改革创新基金项目(J20221191);山西省教育科学“十四五”规划课题基金项目(GH-220170、GH-21175);太原学院院级科研基金项目(21TYKY01)。

摘　　要：针对传统的无源定位机器学习算法存在定位精度低、鲁棒性差等缺点,提出一种基于块稀疏编码的物联网环境下无源定位机器学习算法。引入块稀疏编码策略,利用l范数作为正则项,充分利用无源定位信号的自然群结构实现稀疏解中的群选择;引入近端算子,有效解决提出方法中的非光滑凸优化问题;在原始感知信号中加入严重的高斯噪声,保护网络隐私,提高模型的鲁棒性。实际数据驱动的实验结果表明,该算法在强噪声环境下仍能获得鲁棒的定位和信号恢复性能,优于现有的无源定位方法,验证了该方法的有效性。In view of the shortcomings of traditional machine learning algorithms for passive location,such as low positioning accuracy and poor robustness,a machine learning algorithm for passive location in the internet of things environment based on block sparse coding was proposed.The block sparse coding strategy was introduced,and lnorm was used as the regular term,and the natural group structure of passive location signal was fully utilized to realize the group selection in sparse solution.The near end operator was introduced to solve the non-smooth convex optimization problem effectively.Serious Gaussian noise was added to the original sensing signal,which not only protected the network privacy,but also improved the robustness of the model.Experimental results of real data-driven show that the proposed algorithm can still achieve robust localization and signal recovery performance in strong noise environment,which is better than the existing passive location methods,verifying the effectiveness of the proposed method.

关 键 词：无源定位 机器学习 物联网 块稀疏编码 鲁棒性 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置] TN95[自动化与计算机技术—控制科学与工程]