Optimal state and branch sequence based parameter estimation of continuous hidden Markov model

A parameter estimation algorithm of the continuous hidden Markov model isintroduced and the rigorous proof of its convergence is also included. The algorithm uses theViterbi algorithm instead of K-means clustering used in the segmental K-means algorithm to determineoptimal state and branch sequences. Based on the optimal sequence, parameters are estimated withmaximum-likelihood as objective functions. Comparisons with the traditional Baum-Welch and segmentalK-means algorithms on various aspects, such as optimal objectives and fundamentals, are made. Allthree algorithms are applied to face recognition. Results indicate that the proposed algorithm canreduce training time with comparable recognition rate and it is least sensitive to the training set.So its average performance exceeds the other two.

A Gaussian Multivariate Hidden Markov Model for Breast Tumor Diagnosis

The stage of a tumor is sometimes hard to predict, especially early in its development. The size and complexity of its observations are the major problems that lead to false diagnoses. Even experienced doctors can make a mistake in causing terrible consequences for the patient. We propose a mathematical tool for the diagnosis of breast cancer. The aim is to help specialists in making a decision on the likelihood of a patient’s condition knowing the series of observations available. This may increase the patient’s chances of recovery. With a multivariate observational hidden Markov model, we describe the evolution of the disease by taking the geometric properties of the tumor as observable variables. The latent variable corresponds to the type of tumor: malignant or benign. The analysis of the covariance matrix makes it possible to delineate the zones of occurrence for each group belonging to a type of tumors. It is therefore possible to summarize the properties that characterize each of the tumor categories using the parameters of the model. These parameters highlight the differences between the types of tumors.

Statistical Modeling with a Hidden Markov Tree and High-resolution Interpolation for Spaceborne Radar Reflectivity in the Wavelet Domain

With the increasing availability of precipitation radar data from space,enhancement of the resolution of spaceborne precipitation observations is important,particularly for hazard prediction and climate modeling at local scales relevant to extreme precipitation intensities and gradients.In this paper,the statistical characteristics of radar precipitation reflectivity data are studied and modeled using a hidden Markov tree(HMT)in the wavelet domain.Then,a high-resolution interpolation algorithm is proposed for spaceborne radar reflectivity using the HMT model as prior information.Owing to the small and transient storm elements embedded in the larger and slowly varying elements,the radar precipitation data exhibit distinct multiscale statistical properties,including a non-Gaussian structure and scale-to-scale dependency.An HMT model can capture well the statistical properties of radar precipitation,where the wavelet coefficients in each sub-band are characterized as a Gaussian mixture model(GMM),and the wavelet coefficients from the coarse scale to fine scale are described using a multiscale Markov process.The state probabilities of the GMM are determined using the expectation maximization method,and other parameters,for instance,the variance decay parameters in the HMT model are learned and estimated from high-resolution ground radar reflectivity images.Using the prior model,the wavelet coefficients at finer scales are estimated using local Wiener filtering.The interpolation algorithm is validated using data from the precipitation radar onboard the Tropical Rainfall Measurement Mission satellite,and the reconstructed results are found to be able to enhance the spatial resolution while optimally reproducing the local extremes and gradients.

Quality prediction of batch process using the global-local discriminant analysis based Gaussian process regression model

The conventional single model strategy may be ill- suited due to the multiplicity of operation phases and system uncertainty. A novel global-local discriminant analysis （GLDA） based Gaussian process regression （GPR） approach is developed for the quality prediction of nonlinear and multiphase batch processes. After the collected data is preprocessed through batchwise unfolding, the hidden Markov model （HMM） is applied to identify different operation phases. A GLDA algorithm is also presented to extract the appropriate process variables highly correlated with the quality variables, decreasing the complexity of modeling. Besides, the multiple local GPR models are built in the reduced- dimensional space for all the identified operation phases. Furthermore, the HMM-based state estimation is used to classify each measurement sample of a test batch into a corresponding phase with the maximal likelihood estimation. Therefore, the local GPR model with respect to specific phase is selected for online prediction. The effectiveness of the proposed prediction approach is demonstrated through the multiphase penicillin fermentation process. The comparison results show that the proposed GLDA-GPR approach is superior to the regular GPR model and the GPR based on HMM （HMM-GPR） model.

Comparison of Khasi Speech Representations with Different Spectral Features and Hidden Markov States

In this paper,we present a comparison of Khasi speech representations with four different spectral features and novel extension towards the development of Khasi speech corpora.These four features include linear predictive coding(LPC),linear prediction cepstrum coefficient(LPCC),perceptual linear prediction(PLP),and Mel frequency cepstral coefficient(MFCC).The 10-hour speech data were used for training and 3-hour data for testing.For each spectral feature,different hidden Markov model(HMM)based recognizers with variations in HMM states and different Gaussian mixture models(GMMs)were built.The performance was evaluated by using the word error rate(WER).The experimental results show that MFCC provides a better representation for Khasi speech compared with the other three spectral features.

Hidden Markov Models with Factored Gaussian Mixtures Densities

We present a factorial representation of Gaussian mixture models for observation densities in Hidden Markov Models(HMMs), which uses the factorial learning in the HMM framework. We derive the reestimation formulas for estimating the factorized parameters by the Expectation Maximization (EM) algorithm. We conduct several experiments to compare the performance of this model structure with Factorial Hidden Markov Models(FHMMs) and HMMs, some conclusions and promising empirical results are presented.

基于连续隐Markov模型的发酵过程关键状态变量软测量

针对生物发酵过程中一些关键状态变量难以用仪表进行在线检测的问题,提出了一种基于连续隐Markov模型(CHMM)的软测量方法.首先,为增加模型预测的鲁棒性,采用多观测样本序列训练CHMM模型库,运用Baum-Welch参数重估修正公式实现CHMM参数优化;然后,将新样本观测向量输入训练好的CHMM模型库,并基于Viterbi算法计算新样本在模型库中各个CHMM内的输出概率;最后,通过加权平均运算得出软测量结果.将所提出的方法应用于红霉素发酵过程菌体质量浓度软测量,并进行了建模与仿真研究.结果表明:所建CHMM软测量模型能够很好地用于生物发酵过程菌体质量浓度值预测,且预测效果优于人工神经网络软测量模型.

基于流形学习和隐Markov模型的故障诊断

为实现旋转机械故障诊断的自动化与高精度,提出基于正交邻域保持嵌入和连续隐Markov模型的模型诊断方法。将活动件故障振动信号进行经验模式分解并构造Shannon熵得到高维特征向量,利用正交邻域保持嵌入将高维特征向量约简为低维特征向量,并输入到各个状态连续隐Markov链进行旋转机械的故障模式识别。通过深沟球轴承故障诊断实例验证了该模型的有效性。

基于连续隐Markov模型的理论线损率预测研究

提出一种基于连续隐马尔可夫模型的理论线损率预测方法,运用Baum-Welch参数重估修正公式实现CHMM参数优化,建立CHMM模型。依据CHMM模型概率输出判别各可选仿真模型相对于实际系统的有效性。结果表明,该模型运行速度快,在理论线损率预测中有更高的精度。

Segmentation of MS lesions using entropy-based EM algorithm and Markov random fields

This paper presents an approach for fully automatic segmentation of MS lesions in fluid attenuated inversion recovery (FLAIR) Magnetic Resonance (MR) images. The proposed method estimates a gaussian mixture model with three kernels as cerebrospinal fluid (CSF), normal tissue and Multiple Sclerosis lesions. To estimate this model, an automatic Entropy based EM algorithm is used to find the best estimated Model. Then, Markov random field (MRF) model and EM algorithm are utilized to obtain and upgrade the class conditional probability density function and the apriori probability of each class. After estimation of Model parameters and apriori probability, brain tissues are classified using bayesian classification. To evaluate the result of the proposed method, similarity criteria of different slices related to 20 MS patients are calculated and compared with other methods which include manual segmentation. Also, volume of segmented lesions are computed and compared with gold standard using correlation coefficient. The proposed method has better performance in comparison with previous works which are reported here.

Statistical Model-Based Driving Situation Recognition

The authors propose a two-stage method for recognizing driving situations on the basis of driving signals for application to a safe human interface of an in-vehicle information system. In first stage, an unknown driving situation is determined as stopping behavior or non-stopping behavior. In second stage, a Hidden Markov Model （HMM）-based pattern recognition method is used to model and recognize six non-stopping driving situations. The authors attempt to find the optimal HMM configuration to improve the performance of driving situation recognition. Center for Integrated Acoustic Information Research （CLAIR） in-vehicle corpus is used to evaluate the HMM-based recognition method. Driving situation categories are recognized using five driving signals. The proposed method achieves a relative error reduction rate of 30.9% compared to a conventional one-stage based HMMs.

基于聚类和马尔可夫链的货车实际工况构建方法研究

针对货车法规工况与实际使用工况的油耗等性能相差较大问题,利用车联网平台获取目标样车实车行驶数据,提出基于高斯混合模型聚类和马尔可夫链结合的货车实际工况构建方法。通过主成分分析、高斯混合模型聚类分析将运动学片段分类,类中以速度区间划分行驶状态,采用Laplace Smoothing方法来估计状态转移概率,利用马尔可夫链法合成类内候选工况,选择类内最佳工况,构建目标货车实际行驶工况。结果表明:与原始数据相比,所建工况速度-加速度联合概率密度分布吻合较好,整体特征参数平均相对误差为4.32%;与实测百公里油耗相比,法规工况和所建工况的仿真油耗相对误差分别为26.33%、4.86%。所建工况精度较高,能够有效反映车辆实际行驶特征及油耗性能,可以为车辆的精细化设计和实际道路整车性能评价提供条件和依据。

基于GMM的幅度相位联合编码CVQKD安全性分析

为了提高离散调制连续变量量子密钥分发协议性能,采用幅度相移键控(APSK)联合调制格式方法,在接收端采用高斯混合模型分类算法识别量子态来提升系统的性能。将密钥传输系统分为状态学习和状态预测两个阶段,在状态学习阶段基于高斯混合模型的分类器对已知类别的量子态进行训练,学习不同类别量子态的幅度相位分布情况;在状态预测阶段则采用最小欧氏距离计算出待测量子态属于每个类别的后验概率,从而判定待测量子态的类别,并通过参数估计、反向协调和保密增强生成最终密钥。结果表明,在反向协调和集体攻击下128-APSK离散调制连续变量量子密钥分发协议能够有效生成安全密钥,当安全码率为10-6 bit/symbol时,传输距离可接近60 km。该研究为进一步提高离散调制连续变量量子密钥分发协议的系统性能提供了参考。

基于混合高 斯-隐马尔可夫模型的驾驶意图识别方法研究

驾驶意图的准确识别,能为车辆轨迹预测和行驶风险评估提供有力的理论支撑。首先,提取大量车辆轨迹数据,通过拼接重构方法得到同时包含3类驾驶意图的轨迹数据;其次,以车辆横向速度、加速度以及偏移量作为驾驶意图的表征参数,构建基于混合高斯—隐马尔可夫模型的驾驶意图识别方法,并通过对比不同识别窗口长度得到:窗口长度为2 s时总体精度最高;然后,为避免单点误判对模型精度的影响,设计1种多点识别的修正方法,对驾驶意图结果进一步修正,多次测试发现以连续3帧作为观察窗口时修正效果最好,驾驶意图的识别率高达98.84%。研究成果能应用于轨迹预测和风险评估中,进而为道路交通安全性的提高起到一定的推动作用。

全自动智能垃圾分类机开关的语音控制方法

针对垃圾分类机使用环境噪声复杂导致语音识别质量较低的问题,提出全自动智能垃圾分类机开关的语音控制方法。首先输入语音信息,对语音信号进行预处理,提取信号特征,提取并预处理语音特征;结合隐马尔可夫模型和基于子空间的混合高斯模型构建语音识别模型;在训练完成的模型中输入语音特征,实现全自动智能垃圾分类机开关的语音控制。实验结果表明,所提方法的主观语音质量评估得分大于1.5,识别控制率大于50%,具有较强的环境噪声抑制能力,提高了语音识别质量。

基于CGHMM的轴承故障音频信号诊断方法

轴承音频信号包含其运行状态的重要信息,通过分析这些信息就能对轴承故障进行有效诊断。率先引入基于连续高斯混合密度隐马尔可夫模型的轴承故障音频诊断方法,避免矢量量化带来的数据处理误差,提高了系统诊断精度;引入基于聚类算法的模型参数初始化方法和标定系数的前向-后向算法,简化系统复杂度,加快了训练和诊断速度,进一步提高了诊断精度。实验结果表明,诊断精度达到98.75%,具有很好的应用前景。

基于高斯混合隐马尔科夫模型的高速公路超车行为辨识与分析

基于驾驶模拟器实验数据,结合高斯混合隐马尔可夫模型(GM-HMM),对高速公路的超车行为进行辨识,并对驾驶员意图和超车行为是否正常进行分析。结果表明,基于GM-HMM的辨识方法能有效辨识多种驾驶工况下的不正常超车行为。

基于自回归–连续隐马尔可夫模型的离心泵故障诊断

根据振动与语音信号的相似性和离心泵故障信号的特点,将连续隐马尔可夫模型引入了离心泵的故障诊断中。利用自回归谱不受数据长度的限制,及自回归模型参数对状态变化规律反映敏感的特点,以信号的12阶自回归谱系数为特征矢量,将其输入到各个状态连续隐马尔可夫进行训练,来实现离心泵的故障诊断。为防止数据下溢,引入前向–后向比例因子算法求其对数似然概率,并且采用K-means算法对连续隐马尔可夫进行参数初始化。在给定的观测序列中每一种模型的优化路径通过Viterbi算法实现,用Baum-Welch算法实现参数重估。最后通过2BA-6A离心泵试验系统验证了该方法的有效性。

利用小波域HMC模型进行遥感图像变化检测

传统阈值检测算法都是基于单函数模型进行的,当差异影像分布函数较复杂时检测结果较差.针对这个问题,提出一种基于小波域的隐马尔科夫链模型的遥感图像变化检测算法.将双高斯混合模型与小波变换结合,解决了单函数模型匹配率低的问题,并通过小波变换引入了图像的空间信息,提高了检测精度.利用双高斯混合模型对小波分解后的多层差异影像进行拟合,根据拟合结果判定待检测点类别.对得到的多层初始分割结果,利用隐马尔科夫链模型根据连续最大后验概率融合,得到最终变化检测图.对真实遥感数据集进行实验,证明这种算法可以得到较好的检测结果.

基于公共信标集的高精度射频指纹定位算法

目前基于WiFi射频指纹定位技术有望成为大规模城区室内外全空间定位的首选.针对RSS信号时变特性严重影响WiFi定位精度和鲁棒性的问题,提出了一种基于公共信标集的高精度射频指纹定位算法.该算法把目标定位看成贝叶斯估计问题,通过采用高斯混合模型更加准确地表征复杂训练指纹的信号特征,以及使用基于Markov链的状态转移模型和基于后验概率的自适应网格集选择机制,充分利用目标的历史状态信息和环境布局信息,不仅减少了定位搜索网格空间,而且还抑制了移动过程中不可能发生的位置跳变,提高了定位精度和鲁棒性.实验结果表明,所提定位算法以90%的概率可获得3m以内的定位误差,其定位性能明显优于传统单一高斯模型.