SAE-DSN:一种具有去噪能力的室内定位回归模型  

SAE-DSN:an Indoor Positioning Regression Model with Denoising Ability

作　　者：宋玲[1,2] 王立颖 SONG Ling;WANG Li-ying(School of Computer and Electronic Information,Guangxi University,Nanning 530004,China;Guangxi Key Laboratory of Multimedia Communications and Network Technology,Nanning 530004,China)

机构地区：[1]广西大学计算机与电子信息学院,南宁530004 [2]广西多媒体通信与网络技术重点实验室,南宁530004

出　　处：《小型微型计算机系统》2023年第10期2255-2261,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61762030)资助;广西创新驱动重大专项项目(桂科AA17204017)资助;广西重点研发计划项目(桂科AB19110050)资助;广西创新驱动发展专项项目(桂科AA20302002-3)资助.

摘　　要：Wi-Fi指纹定位技术因其部署成本低、实用性强等优点被广泛应用,但接收信号强度(RSS)的波动给定位带来了一定的挑战.现有的定位方法只考虑了离线阶段指纹数据的去噪问题,而没有考虑在线阶段指纹数据噪声对定位的影响,当在线阶段收集到的指纹数据噪声过大时,将严重影响定位精度.为此,本文提出一种基于稀疏自编码器(SAE)和深度收缩网络(DSN)的室内定位回归模型,该模型使用稀疏自编码器提取指纹数据的鲁棒特征,同时将软阈值算法作为可训练的收缩函数嵌入到全连接网络中,为每个神经元进行去噪处理,有效减少了在线阶段指纹数据噪声的干扰.实验表明,提出的模型能够有效应对在线阶段高斯和非高斯噪声的干扰,相较于其它同类方法在定位精度上有明显提升.Wi-Fi fingerprint positioning technology is widely used because of its low deployment cost and strong practicability.However,the fluctuation of Received Signal Strength(RSS)brings some challenges to positioning.The existing positioning method only considers the denoising problem of fingerprint data in the offline phase,and does not consider the impact of fingerprint data noise in the online phase on the positioning.When the fingerprint data collected in the online phase is too noisy,it will seriously affect the positioning accuracy.Therefore,this paper proposes an indoor positioning regression model based on Sparse AutoEncoder(SAE)and Deep Shrinkage Network(DSN).The model uses Sparse AutoEncoder to extract the robust features of fingerprint data.At the same time,the soft threshold algorithm is embedded into the fully connected network as a trainable shrinkage function to denoise each neuron.It effectively reduces the interference of fingerprint data noise in the online phase.Experiments show that the proposed model can effectively deal with the interference of Gaussian and non Gaussian noise in the online phase,and the positioning accuracy is significantly improved compared with other similar methods.

关 键 词：Wi-Fi指纹 室内定位 RSS 全连接神经网络 软阈值算法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]