一种基于深度自编码器的指纹匹配定位方法  

A Fingerprint Matching Localization Method Based on Deep Auto Encoder

作　　者：刘震宇[1] 李嘉俊 王昆[1] 

机构地区：[1]广东工业大学信息工程学院,广东广州510006

出　　处：《广东工业大学学报》2017年第5期15-21,共7页Journal of Guangdong University of Technology

基　　金：国家自然科学基金青年科学基金资助项目(61401106);广东省产学研专项(2016B090918126);广州市科技计划项目科学研究专项(2014J4100206)

摘　　要：对于浅层结构的室内定位方法难以构造复杂的室内信号模型,从而导致定位精度不高的问题,本文提出利用深度学习方法建立深层非线性学习模型,提高室内定位精度的RFB-SDAE算法.该算法通过约简及预处理去除指纹采集中的冗余信息,经过堆叠式去噪自编码器多层网络的预训练及参数调整,获得较为精确的非线性室内无线信号的模型表示,并利用该模型实现室内的区域定位.实验结果表明,RFB-SDAE相比于堆叠式去噪编码器和单层神经网络提高了定位准确率,且比堆叠式去噪编码器使用更少的运行时间.Indoor localization methods of simple structure are hard to construct a complex indoor signal model,which leads to a poor position result. To solve this problem, a RFB-SDAE algorithm is proposed. The algorithm uses a deep learning structure to establish a deep nonlinear learning model to improve the indoor position accuracy.It removes the redundant information from fingerprint acquisition by the method of simplification and preprocessing. Via the pertaining and parameter adjustment of the networks of stacked denoising auto encoder, it acquires a more exact indoor wireless signal model which is used for indoor localization. The experimental result shows that, compared with stacked denoising auto encoder（SDAE） and single neural network, the RFB-SDAE has improved its accuracy and shortened its execution time.

关 键 词：室内定位 深度学习 自编码器 指纹匹配 约简 

分 类 号：TN911.23[电子电信—通信与信息系统]