一种基于卷积神经网络的室内定位方法  

An Indoor Positioning Method Based on Convolutional Neural Network

作　　者：张丽 董建 孙长智 刘成刚 Zhang Li;Dong Jian;Sun Changzhi;Liu Chenggang(Department of Electronic and Information Engineering,Bozhou University,Bozhou,Anhui 236800,China)

机构地区：[1]亳州学院电子与信息工程系,安徽亳州236800

出　　处：《黑龙江工业学院学报（综合版）》2024年第5期80-87,共8页Journal of Heilongjiang University of Technology(Comprehensive Edition)

基　　金：2022年安徽省大学生创新创业训练计划项目“一种结合卷积神经网络与WKNN的精确定位方法研究”(项目编号:S202212926009);亳州学院质量工程项目“新工科背景下传感器技术课程实践教学体系的构建与实践”(项目编号:2023XJXM052)。

摘　　要：针对实际室内定位场景中的无线接入点信号不稳定引起定位精度低的问题,提出一种基于卷积神经网络的室内定位方法,该方法包括离线阶段和在线阶段,其中离线阶段主要完成对无线接入点信号采集,经过预处理后作为卷积神经网络模型的训练数据。在线阶段利用训练好的模型完成粗定位,估计位置所在的区域,最后利用加权k近邻算法计算精确的位置坐标。通过与SVR、KNN算法对比,结果表明,在二维平面回归定位问题中优于其他算法。According to the problem of low positioning accuracy caused by unstable wireless access point signals in actual indoor positioning scenarios,an indoor positioning method based on convolutional neural network is proposed.The algorithm includes an offline stage and an online stage.The offline stage mainly completes the acquisition of wireless access point signals,which is used as the training data for the convolutional neural network model.The online stage uses the trained model to estimate the area where the location is located,and finally the weighted k-nearest neighbour algorithm is used to calculate the exact position coordinates.By comparing with SVR and KNN algorithms,the results show that it is superior to other algorithms in the two-dimensional planar regression positioning.

关 键 词：卷积神经网络 室内定位 加权k近邻算法 位置指纹算法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TN925[自动化与计算机技术—控制科学与工程]