【作者】 王宇春；

【导师】 景军；

【作者基本信息】 燕山大学 ， 生物医学工程， 2006， 硕士

【摘要】 脉诊在中医理论与临床诊断中都占有很重要的地位,运用数学分析法判别脉象是近代中医学研究的重要课题之一。本文以中医脉象人工智能辩识的研究与开发为背景,对脉象信号的分析和识别做了研究和探讨。传统的脉象图形数学分析法为时域分析法,对时域内脉象图的特征形参数进行分析研究,寻找出了部分中医脉象的参数定义,解释与疾病的关系。本文将小波分析理论应用于脉象信号的分析和处理,讨论了小波变换对脉象信号除噪的效果,利用小波分析所具有良好的时-频同时局部化的能力和对非平稳信号突变点的检测能力对脉象信号进行了分析和时域特征值的提取,并提取了脉象信号的小波变换在不同尺度上的能量这一新的表征脉象的特征参数。在此基础上提出了利用BP神经网络对健康人和脑血管病人脉象信号进行分类,比较了以脉象信号的频谱能量和以其小波变换高频部分d6尺度上的能量为神经网络输入时的训练结果的差异。尽管文中的训练样本有限,但仿真结果表明:对脉象信号的一些特定的特征值(如原始信号的小波变换在不同尺度上的能量),利用神经网络进行识别是一种可行而有效的方法。最后介绍了基于DSP的脉象信号采集系统的设计。更多还原

【Abstract】 Diagnosis based on pulse tracings plays an important part both in theory and in clinical traditional Chinese medicine. Signal processing and pattern recognition of Human Pulse are studied and discussed in this thesis on the basis of the research and development of AI recognition system for Human Pulse. The traditional mathematical analysis of pulse tracing graph is a time domain analysis that found part parameters defining traditional Chinese medicine pulse tracings and interpreted the relation between pulse tracings and disease of entrails. As for the periodicity of pulse tracing signals, frequency domain analysis has been used. Based on the characteristics of Human Pulse, Wavelet Transform (WT) is originally used to process and analysis it. Wavelet analysis has a good qualities both in time domain and frequency domain and is an ideal tool in analyzing unsteady signal,so it is used to detect the singularity of Human Pulse and extract the features of Human Pulse in time-domain. Human Pulse is characterized by a new feature, which is the energy of its wavelet transform in different scales. BP Neural Network is used to classify the Human Pulse between health and Cerebrovascular Disease (CVD) according to its spectra features and new features extracted on d6 by Wavelet Transform. The results show that different inputs of features will lead to different outcomes of pattern recognition. In spite of limited training samples, the method in this thesis is superior to traditional Pattern Recognition methods if choose suitable features to be input-cell such as features extracted by Wavelet Transform. A data processing system base on DSP is designed according to the new characteristics.更多还原