基于HHT和数据挖掘技术的白细胞信号识别研究

发布时间：2018-01-08 10:26

本文关键词：基于HHT和数据挖掘技术的白细胞信号识别研究　出处：《南昌大学》2015年硕士论文　论文类型：学位论文

【摘要】：血细胞信号的获取作为血液分析仪里面的核心技术,目前已经得到了越来越广泛的研究。血细胞信号具有脉冲形状多样、非线性和非平稳的特点,传统方法需要将幅值不变的简谐信号定义为基底,给信号分析带来了诸多限制,导致难以发现血细胞信号内蕴的很多生理或病理特征。因此,研究一种能有效挖掘血细胞信号内在特征的分析方法具有重要研究意义。本文首先对血细胞技术的国内外发展现状及趋势作了介绍,并且重点分析了血细胞信号识别过程中所面临的问题。然后,针对血细胞信号非线性、非平稳的特点,设计了一种基于希尔伯特-黄变换(Hilbert-Huang Transform,HHT)和数据挖掘技术的分类识别方法。阐述了HHT的基本原理和数据挖掘理论。介绍了瞬时频率的概念、HHT的本征模态函数(IMF)和经验模态分解两个关键步骤,以及对Hilbert谱和Hilbert边际谱的定义。同时也说明了数据挖掘的具体过程和一些常用的数据挖掘技术。针对同一IMF分量里面会出现不同时间尺度成分这一现象,本文采用集合经验模态分解方法(EEMD)自适应地分解血细胞信号,获取信号的Hilbert谱、Hilbert边际谱和HHT三维时频谱图。通过对这些谱图数据的分析来提取特征向量：给出所要提取的特征的定义和公式,计算IMF能量与信号总能量的比值、中心频率与强度,并选出其中具有较好区分度的特征值。把这些有效特征值作为分类和预测算法-C4.5决策树的输入信息,结合剪枝技术得到最终的分类模式,并与基于时间域的血细胞信号分类识别进行对比。最后,采用准确率和精度、lift图、ROC曲线、鲁棒性和可解释性等一系列标准对模式进行评估和比对,找出其中的最佳分类模型。研究表明,基于HHT提取特征来构建的分类模型与传统的基于时间域的分类模型相比,区分度明显提高。所以,通过对血细胞信号固有特性的分析,采用HHT算法,解决了其非线性非平稳特点带来的诸多难题,有利于提取和发现血细胞信号的生理或病理特征,给临床医学诊断提供了一种新的思路。
[Abstract]:Blood cell signal acquisition as the core technology of blood analyzer inside, has been extensively studied. Blood cell signal with pulse shapes, nonlinear and non-stationary characteristics, the traditional method of harmonic signal amplitude is constant needs to be defined as substrate, have brought many restrictions in signal analysis, may be difficult to find the intrinsic blood cell signal many physiological or pathological features. Therefore, it has important significance to study an intrinsic characteristic of effective mining blood cell signal analysis method. Firstly, the blood cell technology at home and abroad the status quo and development trend were introduced, and analyzed the face blood cell signal recognition process. Then, the blood cell signal nonlinear, nonstationary characteristics, a design based on Hilbert Huang transform (Hilbert-Huang Transform, HHT) and data mining technology Classification and recognition method of operation. Elaborates the basic theory and principle of HHT data mining. This paper introduces the concept of instantaneous frequency, the intrinsic mode function HHT (IMF) and EMD two key steps, and the definition of the Hilbert spectrum and Hilbert marginal spectrum. It also states the process of data mining and some commonly used data mining technology. For the same IMF components which appear in different time scale components of this phenomenon, this paper adopts the method of ensemble empirical mode decomposition (EEMD) adaptive decomposition of blood cell signal acquisition, signal Hilbert spectrum, Hilbert marginal spectrum and HHT three-dimensional time-frequency spectrum. Based on the analysis of these spectra data to feature extraction: the definition and characteristics of the formula to be extracted by the calculation of the ratio of IMF energy and the total signal energy, center frequency and intensity, and the selected feature which has better discrimination value. These characteristic values as input information classification and prediction algorithm of -C4.5 decision tree based classification model, pruning technique is final, and compared with the classification and recognition of blood cell signal based on time domain. Finally, the accuracy and precision, lift diagram, ROC curve, robustness and a series of standard interpretation etc. on the mode of assessment and comparison, find out the best classification model of them. Research shows that the classification model of HHT to construct the feature extraction and the traditional classification model based on time domain comparison based on discrimination is obviously improved. So, through the analysis on the inherent characteristics of blood cell signal, using HHT algorithm to solve the nonlinear and non many the characteristics of stable problem, is conducive to the extraction and found blood cell signal physiological or pathological features, provides a new way for clinical diagnosis.

【学位授予单位】：南昌大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：R446.11;TP311.13

【参考文献】