高阶神经网络用于心电信号的识别

心电图（ＥＣＧ）的自动识别及分类长期以来一直是较难解决的问题，特别是当截取ＥＣＧ信号的起点不同（相当于信号发生平移）及有基线漂移（相当于信号发生旋转）时，在不作预处理情况下要求仍能正确识别及分类，即要求识别过程具有平移不变性和旋转不变性，更是困扰生物医学工程上作者的困难问题之一。作者采用高阶神经网络（二阶），对五类（正常类、高Ｒ波类、高Ｔ波类、Ｔ波倒置类及心律不齐类）具有平移及旋转的ＥＣＧ信号进行分类识别。结果表明，本网络具有较好的分类效果。平移信号不受任何限制，可任意平移，旋转信号最大旋转幅度为原信号的２０％。本工作全部在微机上完成，而且本网还具有灵敏度高，学习时间短等特点，而有较大的临床实用价值。

Recognition of ECG Using a High Order Neural Network

The classification and recognition of electrocardiogram (ECG) are difficult to solve for a long time, especially when the starting point of selected ECG signal is shifted or the signal is rotated. It. such cases, it requires the recognition process to be translation-invariant and rotation-invariant. The authors introduce a high order (second order) neural network which is used to recognize five types of ECG signals (nomal group. higher QRS wave group, higher T wave group, inversed T wave group and arrhythmia group) with translation and rotation. The results showed that the network is very effective to distinguish the above mentioned 5 types of ECG. The signal could have been translated without any limit ation. The maximum of signal rotation is allowed 20% with respect to the original one. All the task was carried out with microcomputer. Our network is with properties of high sensitivity and short time for learning. It is of great value for clinical use.