基于深度置信网络的室内指纹定位算法  

Indoor finerprint localization algorithm based on deep belif network

作　　者：秦益文 封志宏[1] 苏楠 马丁 李小菲 QIN Yiwen;FENG Zhihong;SU Nan;MA Ding;LI Xiaofei(School of Electric and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China;School of Cyber Security and Computer,Hebei University,Baoding 071002,China;Information Technology Center,Hebei University,Baoding 071002,China)

机构地区：[1]兰州交通大学电子与信息工程学院,甘肃兰州730070 [2]河北大学网络空间安全与计算机学院,河北保定071002 [3]河北大学信息技术中心,河北保定071002

出　　处：《河北大学学报（自然科学版）》2021年第3期321-328,共8页Journal of Hebei University(Natural Science Edition)

基　　金：河北省自然科学基金资助项目(F2019201427);教育部“云数融合科教创新”基金资助项目(2017A20004)。

摘　　要：基于WLAN的指纹识别技术是在每个指纹定位测量来自不同APs的接收信号强度RSS(received signal strength, RSS)以构建指纹.然而,指纹的收集由于环境的变化,需要定期更新指纹库以提高准确性.因此,为减少指纹识别的工作量,提出将深度置信网络算法应用到未标记的RSS测量中,利用生成概率模型来表示不同隐含层的层次隐含特征,提取指纹的隐藏特征,通过预训练和调优阶段,从而在尽可能保持定位精度的前提下减少标记指纹.实验结果表明,本算法在仅使用15%标记指纹时,将定位精度提高了1.876 m.WLAN-based fingerprint recognition technology measures Received Signal Strength(RSS) from different APs at a given location of each fingerprint to construct the fingerprint. However, due to changes in the environment, fingerprint collection needs to be updated regularly to improve accuracy. Therefore, in order to reduce the workload of fingerprint identification, the deep belief network algorithm is applied to the unlabelled RSS measurements, to generate probability model to represent the level of the different hidden layer implied characteristics, to extract the hidden features of fingerprint through preliminary training and tuning phase, and to keep the positioning accuracy under the premise of reducing tag fingerprints as much as possible. Experimental results show that this algorithm improves the positioning accuracy by 1.876 m when only 15% of the fingerprint is used.

关 键 词：指纹定位 深度置信网络 RSS 特征提取 无监督学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]