A Study of ECG and Activity Sensors for Physiological Condition Recognition

  • 江 維鈞

Student thesis: Doctoral Thesis

Abstract

This dissertation presents ECG and activity sensors and algorithms for its application to physiological condition recognition First an ECG arrhythmia classification algorithm is proposed The algorithm extracts the waveform morphology features from ECG signals and then further adopts feature reduction method consisting of principal component analysis (PCA) and linear discriminant analysis (LDA) for selecting significant features The reduced features are sent to a trained probabilistic neural network (PNN) for arrhythmia classification Next a wearable ECG-and-activity sensor system and its sleep stage recognition algorithm are proposed to classify sleep stages by using the combination of physical activity and ECG data In order to improve the accuracy of the classifier the sequential forward selection method is employed to find the significant features at each node of the decision-tree-based support vector machines (DTB-SVMs) classifier The proposed classifier is then used to classify four types of sleep stage and evaluate at-home sleep quality Finally an emotion regulation music player system consisting of music and human emotion detection algorithms is proposed to identify the specific emotion induced by music and customize the music played for different people based on their desired mood The music emotion detection algorithm uses a kernel-based class separability (KBCS) feature selection method and a nonparametric weighted feature extraction (NWFE) method to reduce the dimensions of music features and then a hierarchical SVMs classifier is utilized to detect the music emotion The human emotion detection algorithm uses a genetic algorithm (GA) to select the fittest ECG features for a hierarchical SVMs classifier to detect the human emotion induced by music The experimental results have successfully validated the effectiveness of the proposed ECG and activity sensors and algorithms for its application
Date of Award2014 Dec 8
Original languageEnglish
SupervisorJeen-Shing Wang (Supervisor)

Cite this

'