基於K-means 演算法、小波轉換及支持向量機之心電訊號辨識系統

Abstract

本論文利用小波轉換(Wavelet transform) 、K-means分群法(K-means clustering)及支持向量機(Support vector machine)等方法，建立一個辨識各種心律不整的心電辨識系統。本論文所提的方法可以大致區分為三個階段；第一階段使用K-means分群法把屬於同一類別但相異性卻很大的心律不整訊號分成數個次類別，在每一個次類別，各樣本會有較高的相似性。第二階段則把各次類別裡的每一個心搏樣本利用小波轉換擷取時頻特徵向量。第三階段以每一個心搏樣本的時頻特徵以及形態特徵為訓練資料，並運用支持向量機來建立本辨識系統的模型。為了驗證本系統的有效性以及可靠性，本論文利用MIT-BIH心律不整資料庫進行了三個實驗。實驗的結果本論文所提的方法具有相當高的辨識率達98.2%，最後與各相關辨識系統文獻比較差異。

This paper described an arrhythmia classification system based on the technologies of wavelet transform, k means clustering and support vector machine for the purpose of heartbeat recognition. The method consists of three stages. At the first stage, the waveform of a single heartbeat in each main group is classified into subgroups using k-means clustering technology. At the second stage, the time-frequency features of each heartbeat were extracted by using wavelet transform. At the third stage, the model of the proposed classification system is obtained by using support vector machine (SVM). The training vector of SVM is the combinations of morphological features and time-frequency features extracted using wavelet transform. Three experiments were done to examine the performance and reliability of the proposed classification system. Experiments show that the efficiency and feasibility of this proposed classification system.