基于PARAFAC算法的三维荧光光谱法测定四种混合芳香烃

三维荧光光谱法是描述荧光强度同时随激发波长和发射波长变化的关系谱图,能提供更完整的荧光光谱信息.PARAFAC算法是对不同荧光物质的光谱进行分离,可以从复杂混合成分的三维荧光光谱数据矩阵中将其各自的特征荧光光谱分解出来.将PARAFAC算法与三维荧光光谱法结合,探究了两组分和四组分荧光光谱严重重叠的菲、蒽、芘和荧蒽混合芳烃溶液的分离检测方法,得到了良好的测试效果.     

求解非负矩阵分解的有效集BB梯度算法

非负矩阵分解是在非负限制下的一种将一个高维矩阵分解为两个低维矩阵的分解技术。目前,存在的算法大部分是基于乘性迭代算法和交替最小二乘算法。针对交替最小二乘算法的子问题,文中提出了一种有效集BB梯度法,且该算法是全局收敛的。实验结果显示,该算法比投影梯度算法更为有效。     

求解非负矩阵分解的子空间共轭梯度算法

交替最小二乘法由于其理论可靠性和实际有效性成为非负矩阵分解中备受欢迎的方法之一。文中基于交替最小二乘法将界约束优化中的积极集共轭梯度法运用到非负矩阵分解当中,算法在子问题的求解中,并利用子空间的思想来划分指标集,并利用文献CHENG Wangyou文中的共轭梯度法进行变量更新,在一定条件下证明了新算法的收敛性,实验结果表明算法是有效的。     

基于NMF和LS-SVM的肖像漫画生成算法研究

提出一种基于非负矩阵分解(NMF)和最小二乘支持向量机(LS-SVM)的肖像漫画生成算法.在训练阶段,利用非负矩阵分解来对夸张特征空间数据降维,运用最小二乘支持向量机(LS-SVM)统计学习夸张漫画与人脸之间的关系,建立形状夸张模型.在应用阶段,利用AAM算法提取人脸特征点,形状夸张模型计算出相应的漫画特征点数据,经过图像变形和风格化即可得到最终的肖像漫画.算法实验表明,该算法可以合理地夸张主要特征并避免过度变形.     

四种多环芳烃的三维荧光谱解析

多环芳烃(PAHs)是普遍关注的优先监测污染物,在水环境中其含量很低,但由于其具有高荧光量子产率,可利用荧光法对其进行检测.运用三维荧光光谱法对网种多环芳烃的荧光光谱进行了解析,较二维荧光光谱能更详细地描述PAHs的性质.实验表明t菲与蒽均有三个较强的荧光区域;芘有多个荧光区域;荧蒽有两个荧光区域.菲、蒽、芘和荧蒽的最大荧光峰分别位于250/364 nm、250/402 nm、240/372 nm及2s6/462 nm.选择最大荧光峰值位置对四种多环芳烃做工作曲线,呈现较好的线性关系.该方法为同时检测水源中多组分痕量PAHs提供了方法基础.     

基于约束非负矩阵分解的高光谱图像解混快速算法

约束非负矩阵分解是高光谱图像解混中常用的方法.该方法的求解通常采用投影梯度法,其收敛速度、求解精度和算法稳定性都有待提高.为此,本文针对较优的最小体积约束,提出一种基于约束非负矩阵分解的高光谱图像解混快速算法.首先优化原有的最小体积约束模型,然后设计了基于交替方向乘子法的非凸项约束非负矩阵分解算法,最后通过奇异值分解优化迭代步骤.模拟和实际数据实验结果验证了本文算法的有效性.     

改进的非负矩阵分解语音增强算法

本文提出了一种改进的非负矩阵分解语音增强算法,该算法可分为训练和增强两部分。首先,为了降低训练复杂度,采用卷积非负矩阵分解只提取噪声字典。增强时,考虑语音信号稀疏性比噪声信号稀疏性强,通过稀疏非负矩阵分解重构出语音幅度谱,采用交替方向乘子法进行优化迭代,克服了经典乘性迭代易陷入局部最优、分母只能收敛到零极限等问题。最后,基于算法融合的思想,将重构的语音幅度谱与谱减法、最小均方误差幅度谱估计得到的幅度谱进行加权融合。仿真实验中,在10种不同噪声环境中,通过多种评价标准证明所提算法能取得较好的增强效果。      

基于四阶累积张量方法的欠定盲源信号分离

针对瞬时欠定盲源信号分离问题,提出一种四阶累积张量分解算法.首先构建观察信号四阶累积协方差,依据源信号具有相互独立且均值为零的性质,对累积协方差化简并扩展到张量域,得到四阶累积张量.采用分层交替最小二乘算法对四阶累积张量进行非负库克分解,求得非负库克模型的参数,同时获得非负混合矩阵并求其伪逆,最终估计出源信号.选用真实的语音信号和生物信号进行仿真实验,结果表明该方法提高了源信号和非负混合矩阵的估计性能.     

一种快速非负矩阵分解的人脸识别算法

用于人脸识别的非负矩阵分解算法,虽可提高图像识别率,但因其是通过迭代方法同时计算出基矩阵和系数矩阵,故当迭代次数较多时,计算过程耗时长。文中将二维线性判别分析方法与非负矩阵分解方法融合,提出了一种快速的双边二维非负矩阵分解算法。通过在AR、Ｙale人脸数据库上的实验结果显示,较二维双边非负矩阵分解算法,文中算法不仅使得训练时间大幅减少,而且识别率也有所提高。     

等式约束FIR滤波器设计的投影最小二乘算法

本文考虑具有频域和时域等式约束的FIR滤波器设计问题,提出一个非常有效的新算法——投影最小二乘算法.该算法由两部分组成,前一部分产生一个解析的最小二乘解,后一部分将此解逐次投影到每个等式约束上.该算法有两个显著特点:一是目标函数的Hessian矩阵不要求正定;二是由于采用平方根因子分解来计算增广Hessian矩阵及投影算子矩阵,算法具有很好的数字稳定性.以此算法为核心构成了一个迭代算法,用于实现FIR Nyquist滤波器的minimax设计.设计例子表明了所提算法的有效性和数字稳定性.     

Determination of Luzhou-flavor liquor ages by three-dimensional fluorescence spectroscopy combined with NMF

The determination of Luzhou-flavor liquor ages is carried out by three-dimensional fluorescence spectroscopy combined with non-negative matrix factorization (NMF). 37 samples of aged liquors with weighted ages of 15, 20 and 25 years were prepared by blending three Luzhou-flavor original base liquors with storage ages of 10, 20 and 30 years in different proportions. The fluorescence spectra of the samples were measured, and then factorized into basis matrix and coefficients matrix by multiplicative iterative NMF. The fluorescence spectra, reconstructed from the basis matrix, are similar to the original spectra. The coefficients matrix is taken as the input of support vector machine (SVM) to establish a prediction model for the determination of liquor ages. Compared with the principal component analysis, the prediction model based on SVM has a predicted accuracy better than 91.7%. This method can provide helps for the market supervision on the aged liquors.     

Speaker conversion using kernel non-negative matrix factorization

Voice conversion (VC) based on Gaussian mixture model (GMM) is the most classic and common method which converts the source spectrum to target spectrum. However this method is prone to over-fitting because of its frame-by-frame conversion. The VC with non-negative matrix factorization (NMF) is presented in this paper, which can keep spectrum from over-fitting by adjusting the size of basis vector (dictionary). In order to realize the non-linear mapping better, kernel NMF (KNMF) is adopted to achieve spectrum mapping. In addition, to increase the accuracy of conversion, KNMF combined with GMM (GKNMF) is also introduced into VC. In the end, KNMF, GKNMF, GMM, principal component regression (PCR), PCR combined with GMM (GPCR), partial least square regression (PLSR), NMF correlation-based frequency warping (NMF-CFW) and deep neural network (DNN) methods are compared with each other. The proposed GKNMF gets better performance in both objective evaluation and subjective evaluation.     

In vivo fluorescence spectra unmixing and autofluorescence removal by sparse nonnegative matrix factorization

Fluorescence imaging locates fluorescent markers that specifically bind to targets; like tumors, markers are injected to a patient, optimally excited with near-infrared light, and located thanks to backward-emitted fluorescence analysis. To investigate thick and diffusive media, as the fluorescence signal decreases exponentially with the light travel distance, the autofluorescence of biological tissues comes to be a limiting factor. To remove autofluorescence and isolate specific fluorescence, a spectroscopic approach, based on nonnegative matrix factorization (NMF), is explored. To improve results on spatially sparse markers detection, we suggest a new constrained NMF algorithm that takes sparsity constraints into account. A comparative study between both algorithms is proposed on simulated and in vivo data.     

Morphological decomposition of 2-d binary shapes into modestly overlapped octagonal and disk components.

One problem with several leading morphological shape representation algorithms is the heavy overlapping among the representative disks of the same size. A shape component formed by grouping connected disk centers may use many heavily overlapping disks to represent a simple shape part. Sometimes, these representative disks form complicated structures. A generalized skeleton transform was recently introduced which allows a shape to be represented as a collection of modestly overlapped octagonal shape parts. However, the generalized skeleton transform needs to be applied many times. Furthermore, an octagonal component is not easily matched up with another octagonal component. In this paper, we describe a octagon-fitting algorithm which identifies a special maximal octagon for each image point in a given shape. This transform leads to the development of two new shape decomposition algorithms. These algorithms are more efficient to implement; the octagon-fitting algorithm only needs to be applied once. The components generated are better characterized mathematically. The disk components used in the second decomposition algorithm are more primitive than octagons and easily matched up with other disk components from another shape. The experiments show that the new decomposition algorithms produce as efficient representations as the old algorithm for both exact and approximate cases. A simple shape-matching algorithm using disk components is also demonstrated.     

改进投影梯度非负矩阵分解的单训练样本特征提取研究

人脸识别是当前人工智能和模式识别的研究热点。非负矩阵分解(NMF)能够反映样本的局部的内在的联系,可用于单样本特征提取,但时间复杂度较高。投影梯度(Projected Gradient,PG)优化方法大幅降低了NMF约束优化迭代问题的时间复杂度,但是单训练样本存在对本类信息量描述不足的缺点。为此,该文提出了一种基于改进的投影梯度非负矩阵分解 (Improved Projected Gradient Non-negative Matrix Factorization,IPGNMF) 的单训练样本特征提取方法。在进行PGNMF算子之前,先将训练样本作Gabor分解,分解后的Gabor子图像在各个方向上可以更加丰富的描述样本特征,最后将各个Gabor子图像的PGNMF特征进行融合,作为最终的识别特征。在对人脸库ORL,YEL与FERET的识别实验中,与经典的特征提取方法比较,证明了可以有效地解决单训练样本人脸识别的问题。     

基于集成分量的基因微阵列数据分类方法的研究

构造高精度分类模型是对基因表达谱数据分析的主要研究方向之一,但提取不同特征空间产生的分类效果有很大差异,而集成分类系统在一定程度上提高了分类结果的可靠性和稳定性。构建基于PCA和NMF集成分量系统,并基于分析混合矩阵A的hinton图生物学意义建立集成独立分量选择系统,成功运用到基因表达谱分析,实验结果表明,集成分量分类系统优于单个分类器。     

大规模复杂网络下重叠社区的识别

随着网络规模的不断扩大,经典的复杂网络重叠社识别算法已不能高效处理现有的大规模网络图数据.本文在GraphLab并行计算模型上提出了基于重要节点扩展的重叠社区识别算法DOCVN (Detecting the Overlapping Community algorithm based on Vital Node Expanding in GraphLab).算法选取网络中PageRank值大的节点作为重要节点,计算其他节点归属于重要节点的节点归属度,并以重要节点为中心形成核心社区及扩展社区,最后根据重要节点间的连接紧密度合并核心社区及扩展社区,并计算出每个节点在所属社区里的节点重要度,实现了大规模网络的重叠社区识别.实验表明该算法与PD (Propinquity Dynamics)等现有并行算法相比更能有效地识别大规模网络的重叠社区结构.     

Comparison of multiexcitation fluorescence and diffuse reflectance spectroscopy for the diagnosis of breast cancer (March 2003)

Nonmalignant (n = 36) and malignant (n = 20) tissue samples were obtained from breast cancer and breast reduction surgeries. These tissues were characterized using multiple excitation wavelength fluorescence spectroscopy and diffuse reflectance spectroscopy in the ultraviolet-visible wavelength range, immediately after excision. Spectra were then analyzed using principal component analysis (PCA) as a data reduction technique. PCA was performed on each fluorescence spectrum, as well as on the diffuse reflectance spectrum individually, to establish a set of principal components for each spectrum. A Wilcoxon rank-sum test was used to determine which principal components show statistically significant differences between malignant and nonmalignant tissues. Finally, a support vector machine (SVM) algorithm was utilized to classify the samples based on the diagnostically useful principal components. Cross-validation of this nonparametric algorithm was carried out to determine its classification accuracy in an unbiased manner. Multiexcitation fluorescence spectroscopy was successful in discriminating malignant and nonmalignant tissues, with a sensitivity and specificity of 70% and 92%, respectively. The sensitivity (30%) and specificity (78%) of diffuse reflectance spectroscopy alone was significantly lower. Combining fluorescence and diffuse reflectance spectra did not improve the classification accuracy of an algorithm based on fluorescence spectra alone. The fluorescence excitation-emission wavelengths identified as being diagnostic from the PCA-SVM algorithm suggest that the important fluorophores for breast cancer diagnosis are most likely tryptophan, NAD(P)H and flavoproteins.     

基于稀疏分量分析的振动信号源识别方法

振动传感器接收的信号往往包含不同部件的振动信号和环境噪声,为了从少量振动传感器的接收信号中识别信号源数和各频率分量,提出了一种基于稀疏分量分析的欠定盲源分离方法。该方法首先对混合信号进行时频变换,通过主成分分析提取各个时频点邻域的局部主成分,筛选出单源域特征数据。然后利用余弦距离改进聚类验证技术与模糊聚类算法,对振动源个数进行识别、对聚类参数进行更新,获得信号源数和混合矩阵估计。最后用一系列最小二乘法从混合信号对应的时频点中抽取出源信号。通过仿真实验和实测数据实验验证了本文方法的有效性和稳健性,相比经典时频比方法得到了更稳健、更精确的分离结果,这有助于对机械振动源进行识别和定量评估,以方便后续进行机械状态监测和减振降噪处理。      

Least-square estimation of average power in digital CMOS circuits

The estimation of average-power dissipation of a circuit through exhaustive simulation is impractical due to the large number of primary inputs and their combinations. In this work, two algorithms based on least square estimation are proposed for determining the average power dissipation in complementary metal-oxide-semiconductor (CMOS) circuits. Least square estimation converges faster by attempting to minimize the mean square error value during each iteration. Two statistical approaches namely, the sequential least square (SLS) estimation and the recursive least square estimation are investigated. The proposed methods are distribution independent in terms of the input samples, unbiased and point estimation based. Experimental results presented for the MCNC'91 and the ISCAS'89 benchmark circuits show that the least square estimation algorithms converge faster than other statistical techniques such as the Monte Carlo method and the DIPE