基于转移学习改进低开销指纹的室内定位算法研究

Study of Low Overhead Fingerprint Indoor Location Algorithm Based on Transfer Learning

基于指纹定位算法,为了降低在离线状态目标域的站点测量开销而提高定位的准确性,提出一种通过减少离线训练开销,但是不影响定位精度的基于转移学习(TL)框架的室内定位算法。其基本原理是,根据源域的知识转移重构目标域中的数据分布,使得属于同一个簇的数据在逻辑上彼此接近,而剩余数据则彼此分开。基于TL的框架由2部分组成:度量学习和度量转移,分别用于从源域学习距离度量,并分别为目标域确定最合适的度量。最后利用仿真实验结果证明了基于TL框架定位系统的有效性。

This paper aims to propose a transfer learning （TL） framework based on the extensibility of fingerprint indoor positioning system by reducing off - line training overhead without affecting the positioning accuracy. The basic principle is to reconstruct the data distribution in the target domain according to the knowledge transfer in the source domain, so that the data belonging to the same cluster is logically close to each other, while the remaining data are separated from each other. Specifically, the TL - based framework consists of two parts： measurement learning and measurement transformation, which are used to learn distance measurement from the source domain and determine the most appropriate measures for the target domain respectively. The simulation results show the ef- fectiveness of the system based on transfer learning （TL）.