LDA 기반 분산 레이다 시스템을 활용한 손동작 분류 알고리즘 개발 및 성능 분석

Abstract

In this study, we propose a linear discriminant analysis (LDA)-based classification algorithm for distributed radars that share their data at a data fusion center over wireless channels. Instead of sharing high-dimensional micro-Doppler signature (MDS) data over the resource-limited backhaul, each radar applies the LDA algorithm to reduce the dimension of the data. Because the dimension-reduced MDS data is a vector with complex values, it is expressed as a real-valued probability vector using the softmax processing method to reduce the quantization error. A probability vector with a sum of 1 is transmitted to the fusion center via a pyramid vector quantization (PVQ) scheme. To improve the classification performance at the fusion center, bit resources were adaptively allocated based on the separability of the distributed radars. The proposed algorithm was assessed using MNIST(modified national institute of standards and technology database) data as well as hand gesture data measured using a USRP(universal software radio peripheral)-based test bed.