基于图谱分解的无线定位算法

基于有监督学习的射频指纹定位方法是室内高精度无线定位技术的一个研究热点. 针对有监督学习方法存在训练数据集采集代价较高的问题, 本文提出了一种基于半监督学习的室内无线定位算法. 该算法采用基于Laplacian矩阵谱分解的方法获取训练数据在特征向量空间上的表示, 然后通过有标记数据在特征向量空间上的标记对齐, 实现对未标记数据的标记. 实验结果表明, 仅需少量的有标记数据(20%左右), 便能以较高的精度(80%左右)实现对未标记数据的标记, 从而有效降低了训练开销.

A Wireless Localization Algorithm Based on Spectral Decomposition of the Graph Laplacian

Fingerprint localization based on supervised learning is a hot spot for high-accuracy indoor wireless localization. In order to reduce the training cost of supervised learning method, this paper presents a novel localization algorithm based on semi-supervised learning, which applies spectral decomposition of Laplacian matrix to labeling the unlabeled data through aligning the labeled data in the eigenvectors space. The experimental results show that this algorithm can label the unlabeled data with a high accuracy (about 80%) using only a small amount of labeled data (about 20%), which effectively reduces the data collection cost.