基于局部重构的无监督特征选择方法

利用局部线性嵌入(LLE)算法中获得局部邻域之间的重构关系与使用最小角回归方法解决L1归一化问题都使用回归方法,针对在通过映射获得低维嵌入空间与通过特征选择获得低维空间上有着一致的思想,提出一种能保持局部重构关系的无监督谱特征选择方法.该方法利用最小二乘法计算样本的邻域重构系数,并用这些系数表示样本之间的关系,通过解决稀疏特征值问题获得能够保持样本间关系的低维嵌入空间,最后通过解决L1归一化问题实现自动特征选择.通过在四个不同数据集上的聚类实验结果证明,该方法能更准确地评价每个特征的重要性,能自动适应不同的数据集,受参数影响更小,可以明显提升聚类效果.     

结合Haar与Gabor特征的Adaboost人脸识别改进算法

在模式识别的众多领域中,针对于人脸识别的研究与应用逐渐成为重点和难点。尽管近些年众多学者不断钻研并改进人脸识别的算法,但是在复杂光照条件下和不同肤色的人脸识别中仍然存在着重重的不足。尤其是由于现阶段数据的计算速度和存储条件仍然不能很好地去适应优秀的算法对其的要求,因此如何改进算法,提高人脸识别的精度是本文主要研究的问题。本文结合Haar与Gabor特征提出了Adaboost人脸识别算法的改进方法,提高了人脸识别的速度和精度。     

Determination of Dodecanol and Short‐Chained Ethoxylated Dodecanols by LC–MS/MS (with Electrospray Ionization) After Their Derivatization (with Phenyl Isocyanate)

This report describes the application of LC–MS/MS for the separation of dodecanol (C₁₂OH) and homogenous fatty alcohols ethoxylated (AE) containing a dodecyl moiety and 1–9 ethoxy groups. These ethoxylates and free alcohol were derivatized for LC–MS/MS analysis with phenyl isocyanate (PIC). The derivatives of analytes with PIC were separated using a C18 column. Gradient elution with a mixture of ethyl acetate and acetonitrile (5 mM) was employed. The described determination method is characterized by low detection limits (range from 0.005 µg L^?1 for: C₁₂OH, C₁₂EO_2–7 to 1 µg L^?1 for C₁₂EO₁) and quantification limits (range from 0.01 µg L^?1 for: C₁₂EO_5–7 to 2 µg L^?1 for C₁₂EO₁). The developed and validated method was used in combination with liquid–liquid extraction (using ethyl acetate) in order to identify and quantitatively determine the C₁₂OH and C₁₂EO_1–9 present in environmental samples collected from Warta river water in Poznan.     

LLE-based classification algorithm for MMW radar target recognition

In this paper, a new multiclass classification algorithm is proposed based on the idea of Locally Linear Embedding (LLE), to avoid the defect of traditional manifold learning algorithms, which can not deal with new sample points. The algorithm defines an error as a criterion by computing a sample’s reconstruc-tion weight using LLE. Furthermore, the existence and characteristics of low dimensional manifold in range-profile time-frequency information are explored using manifold learning algorithm, aiming at the problem of target recognition about high range resolution MilliMeter-Wave (MMW) radar. The new algo-rithm is applied to radar target recognition. The experiment results show the algorithm is efficient. Com-pared with other classification algorithms, our method improves the recognition precision and the result is not sensitive to input parameters.     

2-丁酮-水-甲苯、环己酮、4-甲基-2-戊酮三元体系液液平衡的测定及关联

测定了2-丁酮-甲苯-水体系在303.15和323.15K、2一丁酮一4一甲基一2一戊酮一水体系在293.15、303.15和313.15K,2-丁酮-环己酮-水体系在293.15和313.15K下的LLE数据.实验数据用NRTL方程关联,部分互溶对和互溶对的方程参数分别从二元和三元LLE数据得到,关联结果与实验数据符合良好.     

What does the NRTL model look like? Determination of boundaries for different fluid phase equilibrium regions

The adequate correlation of experimental phase equilibrium data by using any thermodynamic model represents a key point for the rigorous design of the chemical processes. However, this data correlation process remains a challenging problem, due to the high nonlinearity of the models used and the consequent problems of convergence. The present work analyzes the nonrandom two-liquids mode (NRTL) model to check the relation between the values of the binary interaction parameters and the corresponding phase equilibria that the NRTL model can or cannot reproduce. Additionally, in the case of the LLE, empirical equations have been established as a function of the nonrandomness parameter, to model the trajectory of the different miscibility boundaries observed. These functions can be used as constraints in the experimental data correlation procedure to reduce the search space and guarantee the consistency of the obtained binary interaction parameters with the experimental behavior of the system under study.     

采用局部线性嵌入的稀疏目标跟踪方法

目标跟踪是从复杂的背景中辨认出运动目标,并且对目标进行准确且连续的追踪。如何在遮挡、形变、背景复杂的条件下鲁棒性跟踪目标仍是亟待解决的问题。针对遮挡和形变问题,提出一种局部线性嵌入(LLE)和稀疏表示的算法来有效的学习外观模板。其中LLE是流形学习的一种典型算法。在该算法中每个点的近邻权值在平移、旋转、伸缩变化下是保持不变的,因此可以用来提取目标的本质特征,发现数据的内在规律。算法首先采用局部线性嵌入提取低维特征,提取后的特征作为基向量与琐碎模板组成稀疏原型,稀疏原型用于模板的更新。算法保持了原有稀疏跟踪方法对遮挡处理的优势,同时对目标形变有较好的稳健性。实验结果表明,跟踪算法比其他7个常用的算法在9个视频序列中有较好的鲁棒性能。     

基于局部超平面的流形奇异值点去除算法

局部线性嵌入算法通常用于高维流形数据降维,具有结构简单、不易陷入局部极小值、能保持局部几何结构不变的特点,但它对噪声和干扰奇异值点非常敏感。为此,提出基于局部超平面的流形奇异值点去除算法,将样本点的邻域投影到超平面空间,使干扰奇异值点投影远离流形样本点投影,而流形样本点投影则表现为聚集特征,同时找出邻域中所有远离聚集中心的样本点作为干扰奇异值点。仿真实验结果验证了该算法的正确性和有效性。     

基于核局部线性嵌入算法的图像去噪方法

利用局部线性嵌入算法进行图像去噪时,如果局部近邻样本呈现非线性关系,图像去噪效果会受到影响。针对该问题,提出基于核局部线性嵌入算法的图像去噪方法。通过非线性核函数将样本映射到高维线性空间,在高维空间运用局部线性嵌入算法进行图像去噪。实验结果表明,该方法能有效地对高维非线性图像进行去噪,性能优于中值滤波算法和局部线性嵌入算法。     

应用相关近邻局部线性嵌入算法的高光谱遥感影像分类

传统的局部线性嵌入（LLE）算法需用欧氏距离度量近邻,但欧氏距离只表示两点间的直线距离,在高维空间中不一定能反映数据间的真实空间分布,导致近邻选取不稳定。针对此问题,本文提出了相关近邻（CN）LLE（CN-LLE）和相关最近邻分类（CNN）算法。提出的算法首先利用相关系数度量数据间的近邻,实现更准确的局部重构,提取鉴别特征;然后用CNN对低维嵌入特征进行分类。在KSC和Indian Pine高光谱遥感数据集上的地物分类实验结果表明：本文提出的CN-LLE+CNN算法比LLE、LLE+CNN和CN-LLE等算法的总分类精度提升了2.11%~11.55%,Kappa系数提升了0.026~0.143。由于该算法增加了近邻为同类的概率,便于更有效地提取同类数据的鉴别特征,且有更好的稳定性,故能更有效地实现高光谱遥感数据的地物分类。