基于上下文感知与自适应响应融合的相关滤波跟踪算法

针对当前相关滤波跟踪算法在抗背景干扰、响应融合方式以及模型更新策略上的不足,提出一种基于上下文感知与自适应响应融合的相关滤波跟踪算法.通过引入上下文感知技术,提高算法在背景杂波及遮挡等跟踪场景下的鲁棒性;通过研究HOG特征和颜色直方图特征二者响应图和响应值的特点,提出一种自适应响应融合方法,提升融合响应图的可靠性;在模型更新方面,采用了高置信度模型更新策略来减轻传统模型更新策略中模型污染及跟踪漂移的问题.实验结果表明,本文算法在OTB50数据集上达到了74.7%的跟踪精度,跟踪成功率为54.8%,均优于对比的主流相关滤波跟踪算法,并且在背景杂波、光照变化、遮挡、运动模糊等复杂跟踪场景中具有较好的跟踪精度与鲁棒性.     

基于灰色关联和D-S证据理论的感应电转子故障诊断

针对单一的灰色关联或D-S证据理论在转子故障诊断中存在的不足,将灰色关联和D-S证据理论相融合的决策级信息融合方法应用到感应电机转子故障诊断中;首先用灰色关联对故障进行初步诊断,然后,将灰色关联分析的诊断输出结果作为D-S证据理论的基本概率分配,最后,依据证据组合规则进行合成,得出转子故障的最终诊断结果;实验仿真结果验证了方法在转子故障诊断中的有效性,可以减小诊断的不确定性,提高故障的诊断准确率和诊断精度。     

Equations of state for hard-sphere fluids

Equations of state and contact values of hard-sphere radial distribution functions (rdf's) which are given by a linear combination of the Percus— Yevick and scaled-particle virial expressions are considered. In the one-component case the mixing coefficient() is, in general, a function of the volume fraction. In mixtures the coefficient( _i ,d _i ), in general, depends upon the volume fraction _i , and diameterd _i , of each species,i andj. For the contact valuesY _ij of the rdf's, the mixing coefficients _ij ( _k ) also depend on speciesi andj. Density expansions for the exact for the one-component hard-sphere fluid are obtained and compared with several approximations made in earlier works and in our own work, as well as with simulations. For a mixture, it turns out that one cannot obtain the exact fourth virial coefficient by using a linear combination of the Percus-Yevick and scaled-particle virial expressions forY _ij unless one allows _ij to depend on mole fractionsx _i even at the zeroth order of its density expansion. We also find that _ij must depend on particle speciesi andj in order to satisfy the exact limits obtained earlier by Sung and Stell. A new equation of state for the binary hard-sphere mixture which satisfies all the exact limits we have considered is suggested.Paper presented at the Tenth Symposium on Thermophysical Properties, June 20–23, 1988, Gaithersburg, Maryland, U.S.A.     

Thermal conductivity of liquid n-alkanes

The thermal conductivity of liquids has been shown in the past to be difficult to predict with a reasonable accuracy, due to the lack of accurate experimental data and reliable prediction schemes. However, data of a high accuracy, and covering wide density ranges, obtained recently in laboratories in Boulder, Lisbon, and London with the transient hot-wire technique, can be used to revise an existing correlation scheme and to develop a new universal predictive technique for the thermal conductivity of liquid normal alkanes. The proposed correlation scheme is constructed on a theoretically based treatment of the van der Waals model of a liquid, which permits the prediction of the density dependence and the thermal conductivity of liquid n-alkanes, methane to tridecane, for temperatures between 110 and 370 K and pressures up to 0.6 MPa, i.e., for 0.3T/T _c0.7 and 2.4P/P _c3.7, with an accuracy of ±1%, given a known value of the thermal conductivity of the fluid at the desired temperature. A generalization of the hard-core volumes obtained, as a function of the number of carbon atoms, showed that it was possible to predict the thermal conductivity of pentane to tetradecane±2%, without the necessity of available experimental measurements.     

非相似度保持投影

由主成分分析(PCA)可知任何一幅人脸图像都可以通过一组特征脸的线性加权来重构,PCA是最小均方误差意义下图像的最优表示,但是传统的PCA最终只通过比较加权系数的欧氏距离来进行识别,没有考虑残差。因此,提出非相似尺度的概念,将两个样本同时投影到相同向量上,在确定它们关系时既考虑投影系数,也考虑重构所产生的残差。两者的投影系数和残差相差越大,说明这两个样本越不相似。和保局投影(LPP)有所不同,非相似度保持投影算法不必预先设定近邻个数,它是利用非相似度的概念,创建非相似度散布矩阵,最终通过最大化目标函数获取最优子空间。在AR库和Feret库上的实验结果证明了该方法的有效性。     

基于人类视觉模型的各向异性扩散滤波方法

针对各向异性扩散模型中扩散系数的梯度阈值确定问题进行了研究,提出了一种基于人类视觉模型的各向异性扩散滤波器。首先在PM方程的基础上研究了各向异性扩散系数中梯度阈值的取值问题。根据仿生学原理,分析了韦伯比曲线,将图像根据背景亮度划分为不同的区域,分别采用不同的公式计算梯度阈值。然后讨论了改进后算法在8邻域内的离散实现问题。仿真结果表明,与传统的恒常梯度阈值扩散模型相比,改进后的算法在有效保留图像重要信息的同时对噪声的抑制效果更为理想。     

基于特征互补图像的人脸识别仿真研究

研究人脸识别精度问题。由于人脸图像中存在大量干扰信息的缺点,而造成了人脸识别正确率下降,为了解决上述问题,提出了一种基于特征互补图像快速特征融合算法。算法通过对人脸图像的位平面切片图像分析,采用位平面图像分解法,通过各种合成策略构造多幅样本图像。并突出高位平面图像,采用两种加权策略将每一幅人脸图像样本都生成"特征互补图像"。然后,直接用图像的二维典型相关分析(2DCCA)法对两种特征互补图像进行特征抽取。最后通过在ORL国际标准人脸库上进行的实验,结果表明,高位平面图像的典型相关鉴别特征提高了正确识别率,并且因为摒弃了原始人脸图像的大部分干扰信息所以具有更强的鲁棒性。     

一种新的红外图像快速分割方法研究

研究红外图像目标分割快速优化问题。在高分辨率红外图像的分割中,红外图像存在数据量大、目标的边缘模糊和噪声较大等导致分辨率低和实时性差。为了快速准确分割,提出基于低尺度分割阈值预测的快速红外图像分割方法(LFIRS)。首先建立尺度阶数计算模型以确定保留原始红外图像目标基本信息所需的最小尺度,在分析多尺度过程对具有目标/背景强相关特性的红外图像进行分割阈值,建立多尺度红外图像分割阈值的相关模型(CLM),结合经典二维阈值法得到的最低两个尺度的阈值CLM模型参数,可以通过CLM模型与最小尺度的快速获取原始尺度的分割阈值,实现红外图像的快速分割。实验结果表明,改进方法提高了图像分辨和分割速度,且改善了分割效果。     

时间相关性经验知识与SVM的融合方法研究

时序数据在时间维度上存在着很强的时间相关性,在时序预测中,利用时序数据的时间相关性特点,构造了一种适用于时序数据预测的时序核函数,实现了将时间相关性融合于支持向量机,并通过人工数据和真实数据验证了时序核函数解决时序预测问题的有效性,并与传统核函数相比具有较好的泛化能力。     

Group contributions to activity coefficients from the hard sphere expansion corresponding states theory

A new method has been developed for predicting liquid activity coefficients in ternary mixtures from group contributions. In this method, activity coefficients are obtained from the excess Gibbs free energy of mixing at constant temperature and pressure. In calculating this excess function, the constituent and mixture Gibbs free energies are each represented by an expansion about a pure reference fluid in powers of ratios of hard-sphere diameters and molecular attraction parameters. When the pure component differs from the reference by a single structural group, these ratios represent, respectively, the size contribution and the attraction contribution of this group to the thermodynamic property of the pure fluid. Contributions of intermolecular repulsion to the excess Gibbs free energy are calculated directly from hard-sphere equations of state for the mixture and pure components. The effect of polar contributions calculated by a Padé approximant is also examined. Results indicate that the method developed from the hard-sphere expansion corresponding-states theory is useful for predicting activity coefficients in ternary mixtures when unlike-pair interaction parameters are fitted to binary activity coefficient data. Furthermore, the method shows promise in providing a theoretical basis for applying group contributions to activity coefficients.Nomenclature A Residual Helmholtz free energy - C _p Constant-pressure heat capacity - H _vap Heat of vaporization - P ^s Saturated vapor pressure - R Gas constant - T Temperature - U ^HS Hard-sphere internal energy - V _c Critical volume - V ^L Liquid volume - x Dimensionless residual thermodynamic property - Z Compressibility factor - d Center-to-center distance between two touching molecules - k Boltzmann constant - n Moles - x Liquid mole fraction - y Vapor mole fraction - Volume expansivity - Liquid activity coefficient - Minimum molecular attraction potential - Isothermal compressibility - Permanent dipole moment - 3.14159... - Density - Fugacity coefficient Paper presented at the Ninth Symposium on Thermophysical Properties, June 24–27, 1985, Boulder, Colorado, U.S.A.