摘要
为了提高T波检测准确率,解决由于反向传播神经元网络(BP神经元网络)连接权值和阈值的初始值选择不合适而导致的无解问题,本研究结合求全局最优解近似值的遗传算法(GA)和求局部最优解精确值的传统BP神经元网络所使用的梯度法,用于T波检测。首先,用GA求得BP神经元网络权值和阈值的全局最优解的近似值;然后,把该近似值作为初始值,训练该神经元网络;最后,用训练好的BP神经元网络识别T波。考虑到T波波峰一定是小波变换(WT)模极大值对,但是反过来,WT模极大值对不一定是T波波峰,首先检测到R波之后一定范围内的WT模极大值对,称为准T波(备选T波),再用训练好的BP神经元网络对刚才检测出的准T波进行筛选和识别,最后确认出真正的T波。MIT数据库实验表明用这样的组合方法检测T波,其检测准确率可达到98%。采取小波变换、GA、BP神经元网络的组合可提高识别T波的正确率。
In order to improve the detection rate of T wave and solve the problem that the back propagate neural network （BP neural network） is invalid when initial weight and threshold values of BP neural network are chosen impertinently, genetic algorithms （GA）＇s was combined with BP＇s gradient method to get whole optimization value and local precision value. After getting an approximation of whole optimization value of weight and threshold values of BP neural network by GA, the approximation was used as the first parameter of BP neural network to train the BP neural network again. The trained BP neural network was then used to recognize T wave of ECG. Considering the fact that the peak of T wave surely corresponds theWavelet Transform （WT）＇s mould max value pairs, but the WT＇s mould max value pairs may not represent the peak of T wave, WT＇ s mould max value pairs, called reserve T wave, were detected firstly in some range after R wave, and the reserve T waves were selected and recognized with the help of the educated BP neural network, lastly, real T waves were confirmed. Experiment results shown that this combining method was useful and applicable for detecting T wave, and the accuracy for the detection of T wave was above 98% for the MIT database It is concluded that the proposed method makes BP neural network to recognize T wave better.
出处
《中国生物医学工程学报》
CAS
CSCD
北大核心
2008年第4期481-484,489,共5页
Chinese Journal of Biomedical Engineering
基金
国家自然科学基金(60373098)
华北科技学院博士基金(A06006)
中国博士后科学基金(第27批)
关键词
T波检测
全局最优解
遗传算法
反向传播神经元网络
小波变换
T wave detection, global optimization value, genetic algorithms, back propagation neural network,wavelet transform
