| Summary: | 碩士 === 國立成功大學 === 電機工程學系碩博士班 === 97 === This study presents a physiological recognition approach based on HRV trends. The focuses of this study include: 1) the development of a user-friendly interface for generating the trends of each HRV parameter and 2) the development of recognition strategies based on the HRV trends. The proposed user-friendly interface enables users to load polysomnography (PSG) signals, to select which channels to display, to specify a time interval, to execute a long-term HRV analysis program, and to inspect the trends of each HRV parameter. The developed recognition strategies include a HRV-trend-based recognition strategy and a HRV-parameter-based recognition strategy. Both strategies consist of the following processes: 1) feature generation, 2) feature selection, 3) feature extraction, and 4) classifier construction for recognition. In the feature generation processes, the trend-based strategy computes statistical features from HRV trends, while the parameter-based strategy calculates features from five-minute HRV analysis results. In the feature selection process, both strategies adopt the best individual N (BIN) as the search strategy and the kernel-based class separability (KBCS) as the selection criterion. Sequentially, principal component analysis (PCA) and linear discriminant analysis (LDA) are adopted in the feature extraction process. Finally, a k-nearest neighbor (k-NN) algorithm is used for the recognition.
The feasibility of these two recognition strategies is verified by two applications: 1) driving condition recognition and 2) severity recognition of Parkinson’s disease. The simulation results demonstrate that both proposed strategies can achieve satisfactory recognition rates in these two applications. In addition, this study compares the average recognition rates of the recognition methods with different feature extraction processes. The results show that the feature extraction process or feature selection process has respective physical meaning in the proposed strategies.
|
|---|
