國立東華大學應用數學系

應    數    講    座

主講人：吳建銘教授
國立東華大學應用數學學系

講  題：Two-lead fetal ECG separation based on nonlinear convolution modeling of Potts Perceptrons

時  間：102 年 04 月 03 日 (星期三)  11:20-11:50

地  點：理工一館 A324 會議室

摘   要

Fetal ECG separation aims to extract source signals oriented from activities of fetal heart subject

to multi-channel maternal observations. Typical maternal ECGs contain eightchannel observations,

which have been processed by unsupervised learning in the field of neural networks for blind

separation of fetal ECGs. This talk presents independent component analysis based on annealed

KLD minimization and demonstrates its applicability for fetal ECG separation. Blind separation of

fetal ECGs is further explored based on supervised learning following the idea of nonlinear

convolution modeling, extended from linear convolution, which contains cascaded linear and post-nonlinear

structures. Potts perceptrons have been shown more general than typical perceptrons.

   Each Potts perceptron has K internal states. When K=2, it reduces to the two-state perceptron.

  Neural organization of multilayer Potts perceptrons and LM (Levenberg-Marquardt) learning have

  been recently proposed. Multilayer Potts perceptrons are applied for nonlinear convolution

  modeling and resolving fetal ECG separation subject to two-lead observations respectively

  measured nearby physical positions of maternal and fetal hearts. Numerical simulations of fetal

  ECG separation subject to two-channel observations are given to verify the effectiveness and

  reliability of the proposed nonlinear convolution modeling.

※※※   歡      迎      參      加   ※※※se1020403

附檔：