利用期望-最大化算法实现基于动态词典的压缩感知

在现有压缩感知(CS)理论中,重构信号需要预设其稀疏表示词典。对于以参数化模型表示的信号,只能预知该词典为某种形式的参数化词典,参数的具体取值难以确定。若将参数设定为取值空间的均匀离散格点,预设词典与真实词典之间的失配将使传统CS重构方法的性能严重恶化。为解决这一问题,该文提出一种基于动态词典的CS重构方法。通过迭代地优化词典参数,该方法在信号重构过程中对词典进行动态调整。为同时实现稀疏恢复与词典调整,该方法利用变分期望-最大化(EM)算法交替执行信号系数估计与词典参数优化。实验结果表明所提方法是有效的。

基于EM的模糊-粗糙集最近邻算法

针对由类的重叠引起的训练样本模糊不确定性,以及属性不足引起的类边界粗糙不确定性,提出一种基于期望-最大化(EM)的模糊-粗糙集最近邻分类算法——EM-FRNN。利用UCI数据库的突发性水污染事件案例进行实验,实验结果表明,与朴素的KNN、模糊最近邻算法、模糊粗糙最近邻算法相比,该算法的运算精度高且计算成本较低。

EM-based detection scheme for differential unitary space-time modulation

The performance loss of an approximately 3 dB signal-to-noise ratio is always paid with conventional differential detection compared to the related coherent detection. A new detection scheme consisting of two steps is proposed for the differential unitary space-time modulation （DUSTM） system. In the first step, the data sequence is estimated by conventional unitary space-time demodulation （DUSTD） and differentially encoded again to produce an initial estimate of the transmitted symbol stream. In the second step, the initial estimate of the symbol stream is utilized to initialize an expectation maximization （EM）-based iterative detector. In each iteration, the most recent detected symbol stream is employed to estimate the channel, which is then used to implement coherent sequence detection to refine the symbol stream. Simulation results show that the proposed detection scheme performs much better than the conventional DUSTD after several iterations.

Real-time reliability evaluation based on damaged measurement degradation data

A method was proposed to evaluate the real-time reliability for a single product based on damaged measurement degradation data.Most researches on degradation analysis often assumed that the measurement process did not have any impact on the product's performance.However,in some cases,the measurement process may exert extra stress on products being measured.To obtain trustful results in such a situation,a new degradation model was derived.Then,by fusing the prior information of product and its own on-line degradation data,the real-time reliability was evaluated on the basis of Bayesian formula.To make the proposed method more practical,a procedure based on expectation maximization (EM) algorithm was presented to estimate the unknown parameters.Finally,the performance of the proposed method was illustrated by a simulation study.The results show that ignoring the influence of the damaged measurement process can lead to biased evaluation results,if the damaged measurement process is involved.

Online split-and-merge expec tation-maximization training of Gaussian mixture model and its optimization

This paper presents a new online incremental training algorithm of Gaussian mixture model （GMM）, which aims to perform the expectation-maximization（EM） training incrementally to update GMM model parameters online sample by sample, instead of waiting for a block of data with the sufficient size to start training as in the traditional EM procedure. The proposed method is extended from the split-and-merge EM procedure, so inherently it is also capable escaping from local maxima and reducing the chances of singularities. In the application domain, the algorithm is optimized in the context of speech processing applications. Experiments on the synthetic data show the advantage and efficiency of the new method and the results in a speech processing task also confirm the improvement of system performance.

隐马尔科夫模型在三维模型自动分类中的应用

针对三维模型的分类问题,提出了一种基于隐马尔科夫模型(HMM)和最大期望(EM)算法的三维模型自动分类方法。将HMM引入三维模型自动分类问题中使得更多先验知识在分类过程中被利用。算法首先对三维模型进行预处理和组合切分,并提取各切分部分的形状直方图特征。对形状直方图特征进行离散归一化后形成HMM模型在某一时刻的观测值,这些观测值将用来训练HMM参数。HMM参数通过EM算法进行估计。最后通过计算未知模型和各类模型的HMM参数间的最大后验概率,获得三维模型的分类结果。在HMM建模过程中利用HMM本身所具有的时序性来描述三维模型的空间几何结构和局部几何特征。实验表明该方法在三维模型自动分类中有较高的准确率。

面向协同过滤的真实偏好高斯混合模型

针对协同过滤问题,提出了一种基于高斯混合的概率模型,称为真实偏好高斯混合模型.用户对项目的评分由三个因素决定:用户对项目主题和内容的真实偏好,用户的评分习惯,以及项目的公众评价.引入了两个隐含变量,分别用于描述用户类和项目类,用户和项目依概率可以同时属于多个类.模型包括离线建模过程和在线预测过程,在线预测可以在常数时间内完成.实验表明新模型的预测结果明显优于其他几种协同过滤算法.

稀疏贝叶斯学习框架下的扩展目标雷达关联成像

传统的关联成像方法未考虑复杂扩展目标的结构信息,在高分辨成像时的应用受到限制,为此提出一种自适应结构配对稀疏贝叶斯学习方法。该方法在稀疏贝叶斯学习的框架内针对扩展目标建立一种结构配对层次化高斯先验模型,然后采用变分贝叶斯期望-最大化算法交替进行目标重构和参数优化。该方法将某一信号分量的重构与周围信号分量联系起来,并能在迭代过程中自适应地调整表征各信号分量相关性的参数。实验结果表明,该方法针对扩展目标可以有效地进行高分辨成像。

基于因子分析的隐马尔可夫模型(英文)

状态输出概率密度为对角协方差矩阵高斯分布的隐马尔可夫模型(HMM-DG)在帧内特征相关建模方面存在缺陷.本文将因子分析方法与HMM-DG的混合高斯建模相结合,提出了一种具有弹性的帧内特征相关隐马尔可夫模型框架——基于因子分析的隐马尔可夫模型(HMM-FA),并导出了HMM-FA的训练算法.理论分析和仿真实验都表明:在训练数据相同的条件下,HMM-FA的性能优于HMM-DG.

基于图像序列的人体跟踪

由于人体的非刚体性和人体之间会经常发生遮挡,使得人体跟踪是一个很有挑战性的课题。针对这一特点,提出了用结合卡尔曼滤波和贝叶斯的方法来完成多个人体的跟踪,先建立简单的背景模型,然后用背景差分法得到前景区域,提取运动人体,并用EM(期望-最大化)算法建立相应的人体模型。在人体间没有发生遮挡时,用卡尔曼滤波方法来跟踪各个人体;人体间出现遮挡时,用贝叶斯方法来判别和跟踪相应人体。实验表明,该方法既能保证跟踪的快速性,又能很好地处理人体间相互遮挡的情况,该算法鲁棒性好,跟踪结果令人满意。

大数据背景下考虑删失特点的继保设备运行状态评估

未能记录到设备失效准确时间的失效数据称为删失数据。针对大数据背景下继保设备失效数据存在删失的特点,提出了大数据背景下考虑删失数据的继保设备运行状态评估方法。首先,通过对继保设备失效数据进行特点分析,对失效数据进行预处理,利用期望-最大化(EM)算法并结合指数分布和威布尔分布模型,估计继保设备失效模型的参数。其次,将参数估计值代入失效模型中,得到设备的可靠度、故障概率密度、失效率、平均无故障时间等可靠性指标。然后,通过仿真验证对比分析偶然失效期和老化失效期内不同估计方法得到的模型参数精度,验证了文中方法处理删失数据的有效性。最后,以某型号继保设备为例分析可靠性指标,验证了利用文中方法规划设备检修周期的可行性。

EFFECTIVE IMAGE SEGMENTATION FRAMEWORK FOR GAUSSIAN MIXTURE MODEL INCORPORATING LOCAL INFORMATION

A new two-step framework is proposed for image segmentation. In the first step, the gray-value distribution of the given image is reshaped to have larger inter-class variance and less intra-class variance. In the sec- ond step, the discriminant-based methods or clustering-based methods are performed on the reformed distribution. It is focused on the typical clustering methods-Gaussian mixture model （GMM） and its variant to demonstrate the feasibility of the framework. Due to the independence of the first step in its second step, it can be integrated into the pixel-based and the histogram-based methods to improve their segmentation quality. The experiments on artificial and real images show that the framework can achieve effective and robust segmentation results.

OFDM blind channel estimation based on polynomial models

A two-dimensional （2-D） polynomial regression model is set up to approximate the time-frequency response of slowly time-varying orthogonal frequency-division multiplexing （OFDM） systems. With this model the estimation of the OFDM time-frequency response is turned into the optimization of some time-invariant model parameters. A new algorithm based on the expectation-maximization （EM） method is proposed to obtain the maximum-likelihood （ML） estimation of the polynomial model parameters over the 2-D observed data. At the same time, in order to reduce the complexity and avoid the computation instability, a novel recursive approach （RPEMTO） is given to calculate the values of the parameters. It is further shown that this 2-D polynomial EM-based algorithm for time-varying OFDM （PEMTO） can be simplified mathematically to handle the one-dimensional sequential estimation. Simulations illustrate that the proposed algorithms achieve a lower bit error rate （BER） than other blind algorithms.

Gaussian mixture model clustering with completed likelihood minimum message length criterion

An improved Gaussian mixture model （GMM）- based clustering method is proposed for the difficult case where the true distribution of data is against the assumed GMM. First, an improved model selection criterion, the completed likelihood minimum message length criterion, is derived. It can measure both the goodness-of-fit of the candidate GMM to the data and the goodness-of-partition of the data. Secondly, by utilizing the proposed criterion as the clustering objective function, an improved expectation- maximization （EM） algorithm is developed, which can avoid poor local optimal solutions compared to the standard EM algorithm for estimating the model parameters. The experimental results demonstrate that the proposed method can rectify the over-fitting tendency of representative GMM-based clustering approaches and can robustly provide more accurate clustering results.

基于块稀疏贝叶斯学习的跳频通信梳状干扰抑制

梳状干扰是对跳频(FHSS)通信的一种有效干扰样式,抑制梳状干扰对于确保FHSS通信的有效性至关重要。现有基于奈奎斯特采样定理的梳状干扰抑制方法存在应用中受限于采样率较高的问题。将压缩感知(CS)应用于FHSS通信梳状干扰的抑制,利用FHSS信号与梳状干扰的不同压缩域特性以及梳状干扰在频域表现出的块稀疏特性,建立了基于块稀疏贝叶斯学习(BSBL)框架的FHSS通信梳状干扰抑制模型。利用期望最大化(EM)算法,设计了基于BSBL_EM的FHSS通信梳状干扰抑制算法。该算法利用BSBL_EM算法从压缩采样数据中重构出梳状干扰,进而在时域对消干扰。仿真结果表明:所提方法能够有效地抑制FHSS通信中的梳状干扰,与传统方法相比具有显著优势,干扰抑制效果受干扰强度、干扰梳齿带宽和压缩率变化的影响;相同干扰强度条件下,梳齿带宽越窄,压缩率越大,干扰抑制效果越好。

Efficient Recovery of Structured Sparse Signals via Approximate Message Passing with Structured Spike and Slab Prior

Due to limited volume, weight and power consumption, micro-satellite has to reduce data transmission and storage capacity by image compression when performs earth observation missions. However, the quality of images may be unsatisfied. This paper considers the problem of recovering sparse signals by exploiting their unknown sparsity pattern. To model structured sparsity, the prior correlation of the support is encoded by imposing a transformed Gaussian process on the spike and slab probabilities. Then, an efficient approximate message-passing algorithm with structured spike and slab prior is derived for posterior inference, which, combined with a fast direct method, reduces the computational complexity significantly. Further, a unified scheme is developed to learn the hyperparameters using expectation maximization(EM) and Bethe free energy optimization. Simulation results on both synthetic and real data demonstrate the superiority of the proposed algorithm.

Improved dark channel image dehazing method based on Gaussian mixture model

To solve the problem of color distortion after dehazing in the sky region by using the classical dark channel prior method to process the hazy images with large regions of sky,an improved dark channel image dehazing method based on Gaussian mixture model is proposed.Firstly,we use the Gaussian mixture model to model the hazy image,and then use the expectation maximization(EM)algorithm to optimize the parameters,so that the hazy image can be divided into the sky region and the non-sky region.Secondly,the sky region is divided into a light haze region,a medium haze region and a heavy haze region according to the different dark channel values to estimate the transmission respectively.Thirdly,the restored image is obtained by combining the atmospheric scattering model.Finally,adaptive local tone mapping for high dynamic range images is used to adjust the brightness of the restored image.The experimental results show that the proposed method can effectively eliminate the color distortion in the sky region,and the restored image is clearer and has better visual effect.

An evolutionary particle filter based EM algorithm and its application

In this paper, an evolutionary recursive Bayesian estimation algorithm is presented, which incorporates the latest observation with a new proposal distribution, and the posterior state density is represented by a Gaussian mixture model that is recovered from the weighted particle set of the measurement update step by means of a weighted expectation-maximization algorithm. This step replaces the resampling stage needed by most particle filters and relieves the effect caused by sample impoverishment. A nonlinear tracking problem shows that this new approach outperforms other related particle filters.

基于Box-Cox变换与随机系数回归模型的非线性退化设备剩余寿命预测方法

准确预测退化设备的剩余寿命可以为设备维护管理提供重要信息支撑,进而避免设备运行中发生计划外失效、减少设备运行维护成本。针对工程实际设备广泛存在的非线性退化现象,提出了基于Box-Cox变换与随机系数回归模型的非线性退化设备剩余寿命预测方法。首先,采用Box-Cox变换对非线性退化数据进行线性化处理,在此基础上通过随机系数回归模型构建退化模型,并运用Bayesian理论与蒙特卡洛-期望最大化算法在线更新模型参数;然后,基于随机系数回归模型的特性,推导出剩余寿命的分布函数以及其点估计值;最后,通过数值仿真和锂电池实际退化数据验证所提方法的有效性。

Maximum Likelihood Blind Separation of Convolutively Mixed Discrete Sources

In this paper,a Maximum Likelihood(ML) approach,implemented by Expectation-Maximization(EM) algorithm,is proposed to blind separation of convolutively mixed discrete sources.In order to carry out the expectation procedure of the EM algorithm with a less computational load,the algorithm named Iterative Maximum Likelihood algorithm(IML) is proposed to calculate the likelihood and recover the source signals.An important feature of the ML approach is that it has robust performance in noise environments by treating the covariance matrix of the additive Gaussian noise as a parameter.Another striking feature of the ML approach is that it is possible to separate more sources than sensors by exploiting the finite alphabet property of the sources.Simulation results show that the proposed ML approach works well either in determined mixtures or underdetermined mixtures.Furthermore,the performance of the proposed ML algorithm is close to the performance with perfect knowledge of the channel filters.