Two-dimensional discriminant transform for face recognition

This paper develops a new image feature extraction and recognition method coined two-dimensional linear discriminant analysis (2DLDA). 2DLDA provides a sequentially optimal image compression mechanism, making the discriminant information compact into the up-left corner of the image. Also, 2DLDA suggests a feature selection strategy to select the most discriminative features from the corner. 2DLDA is tested and evaluated using the AT&T face database. The experimental results show 2DLDA is more effective and computationally more efficient than the current LDA algorithms for face feature extraction and recognition.     

On the estimation of probability of error

This paper considers the problem of estimation of classification error in pattern recognition. A theorem is presented to obtain the changes in the eigenvalues and eigenvectors of matrices of the form S₂⁻¹ S₁, when there are changes of first order of smallness in the real symmetric matrices S_i, I = 1, 2. Based on this theory a computational algorithm is developed for the estimation of classification error of Fisher classifier, using leaving groups out method.     

露天采矿爆破振动对砌体结构破坏效应预测的Fisher判别模型及应用

 应用Fisher判别分析理论并结合工程实际特点,从爆破振动特征参量和砌体结构自身特性这2个方面出发,选取峰值质点振动速度(PPV)、爆破振动主频率、主频率持续时间、灰缝强度、圈梁构造柱、房屋高度、屋盖形式和砖墙面积率8个影响因素作为判别因子,建立爆破振动对砌体结构破坏效应预测的Fisher判别分析模型。将该方法应用到湖北一露天采场爆破振动对砌体结构破坏效应预测问题中,利用现场实测的108组数据进行训练和检验,回判估计的误判率为0.083,通过求解判别函数,认为峰值质点振动速度为最重要的判别指标,其后依次为圈梁构造柱、屋盖形式、砖墙面积率、房屋高度、爆破振动主频率、主频率持续时间和灰缝强度,可以为同类工程在选取爆破振动对砌体结构破坏效应的判别指标方面提供参考。利用其他12组现场数据作为预测样本进行测试,预测结果与实际情况吻合较好。研究表明,该方法回判估计的误判率低,判别性能良好,是爆破振动对砌体结构破坏效应预测的一种有效新方法,可以在实际工程中进行推广应用。     

Biomass-derived CO2 rich syngas conversion to higher hydrocarbon via Fischer-Tropsch process over Fe–Co bimetallic catalyst

Biomass-derived syngas (CO₂ + CO + H₂) has emerged as a potential non-fossil fuel source to yield transportation fuel via Fischer Tropsch Synthesis (FTS) reaction. Thus, the present study demonstrates the conversion of CO₂ containing syngas into fuel range hydrocarbon via Fischer Tropsch Synthesis over Fe–Co bimetallic catalyst. The experimental tests were carried out in a fixed bed continuous reactor to investigate the effect of CO₂ on CO/CO₂ conversion. Accordingly, obtained data were validated by FTS kinetic model for a plug flow reactor. It was found that the unique combination of Fe and Co bimetallic catalyst facilitates both FTS and water gas shift (WGS) reaction simultaneously that helps to convert CO₂ along with CO. It was also observed that the presence of iron in the catalyst helps in conversion of CO₂ into hydrocarbons, only when a particular concentration of CO₂ in syngas is reached, i.e., critical ratio R_C (CO₂/CO + CO₂) due to the occurrence of reverse water gas reaction (RWGS) which varies with the temperature and the feed gas composition (H₂/CO/CO₂ molar ratio). At 240 °C and hydrogen deficient condition, the critical ratio was measured to be 0.74 whereas for hydrogen balanced condition, it was measured 0.6. The kinetic model developed in the present study predicted trends for % CO conversion, % carbon conversion, and % CO₂ conversion which is applicable for a wide range of critical ratio R_C (CO₂/(CO + CO₂) = 0 to 1). The model also predicted that a positive conversion of CO₂ could be achieved at lower CO₂ concentration by increasing the reaction temperature. At 260 °C and 280 °C, the value of Rc were 0.31 and 0.18 were measured.     

基于改进随机森林算法的停电敏感用户分类

目前,我国电网企业对于识别停电投诉风险,开展用户停电敏感程度分析的研究工作还处在起步阶段.为了有效地分析停电用户的敏感程度,提出了一种基于改进随机森林算法的停电敏感用户分类算法.首先,对原始数据进行清洗、特征选择等预处理;接着,采用SMOTE算法增加少数敏感用户样本数据量,解决数据分布不均匀问题;然后,以Fisher比作为特征的重要性度量,按比例随机采样选取具有代表性的特征构成子特征空间;最后,利用随机森林算法识别停电敏感用户.通过在真实停电数据上的实验,验证了提出的方法不仅具有较好的准确性和时间性能,而且可以有效处理高维、冗余特征的数据.     

基于特征波段-Fisher-K近邻的木器漆拉曼光谱的快速无损鉴别

为了实现木器漆的快速无损检测以及精确识别与分类,采集并获取了晨阳等3种品牌木器漆样本的拉曼光谱,并考察了基线校正、Savitzky-Golay九点平滑法、一阶导数和二阶导数等不同预处理方法的处理效果,建立了特征波段比值、Fisher判别、K近邻(KNN)模型。结果表明:特征波段比值法能以1358cm-1/1239cm-1表征3种木器漆的特征;基于Fisher判别的基线校正、平滑和二阶导数处理的拉曼光谱模型的分类准确率最高,能实现100%区分;在相同的预处理下,KNN判别模型的准确率仅为88.5%。基于二阶导数的拉曼光谱结合特征波段-FisherKNN法能为不同品牌木器漆的准确检测提供一种新的快速无损分析手段,具有普适性和一定的借鉴意义。     

Relation between quantum probe and entanglement in n-qubit systems within Markovian and non-Markovian environments

We address in detail the process of parameter estimation for an n-qubit system dissipating into a cavity in which the qubits are coupled to the single-mode cavity field via coupling constant g which should be estimated. In addition, the cavity field interacts with an external field considered as a set of continuum harmonic oscillators. We analyse the behaviour of the quantum Fisher information (QFI) for both weak and strong coupling regimes. In particular, we show that in strong coupling regime, the memory effects are dominant, leading to an oscillatory variation in the dynamics of the QFI and consequently information flowing from the environment to the quantum system. We show that when the number of the qubits or the coupling strength rises, the oscillations, signs of non-Markovian evolution of the QFI, increase. This indicates that in the strong-coupling regime, increasing the size of the system or the coupling strength remarkably enhances the reversed flow of information. Moreover, we find that it is possible to retard the QFI loss during the time evolution and therefore enhance the estimation of the parameter using a cavity with a larger decay rate factor. Furthermore, analysing the dynamics of the QFI and negativity of the probe state, we reveal a close relationship between the entanglement of probes and their capability for estimating the parameter. It is shown that in order to perform a better estimation of the parameter, we should avoid measuring when the entanglement between the probes is maximized.     

基于Mask R-CNN改进的遥感图像舰船检测

遥感图像中舰船朝向不确定性,舰船种类的多样性以及和其他海上及港口物体之间的相似性,使舰船检测的性能下降严重。针对这一问题,使用一种简单且有效的方法来训练有旋转不变性和Fisher判别的Mask R-CNN舰船检测模型,通过优化模型的目标函数以提高舰船检测性能,在保持原有检测模型结构不变的基础上引入两个正则化器,第一个正则化器加强训练样本旋转之前和之后的特征联系,第二个正则化器限制卷积神经网络有小的类内散度和大的类间散度。实验中,在Kaggle遥感图像船只检测数据集上验证了所提出的方法提高了检测遥感图像中舰船目标的性能。     

非贪婪的鲁棒性度量学习算法

目的 度量学习是机器学习与图像处理中依赖于任务的基础研究问题。由于实际应用背景复杂,在大量不可避免的噪声环境下,度量学习方法的性能受到一定影响。为了降低噪声影响,现有方法常用L1距离取代L2距离,这种方式可以同时减小相似样本和不相似样本的损失尺度,却忽略了噪声对类内和类间样本的不同影响。为此,本文提出了一种非贪婪的鲁棒性度量学习算法——基于L2/L1损失的边缘费歇尔分析（marginal Fisher analysis based on L2/L1 loss,MFA-L2/L1）,采用更具判别性的损失,可提升噪声环境下的识别性能。方法 在边缘费歇尔分析（marginal Fisher analysis,MFA）方法的基础上,所提模型采用L2距离刻画相似样本损失、L1距离刻画不相似样本损失,同时加大对两类样本的惩罚程度以提升方法的判别性。首先,针对模型非凸带来的求解困难,将目标函数转为迭代两个凸函数之差便于求解;然后,受DCA（difference of convex functions algorithm）思想启发,推导出非贪婪的迭代求解算法,求得最终度量矩阵;最后,算法的理论证明保证了迭代算法的收敛性。结果 在5个UCI（University of California Irrine）数据集和7个人脸数据集上进行对比实验：1）在不同程度噪声的5个UCI数据集上,MFA-L2/L1算法最优,且具有较好的抗噪性,尤其在30%噪声程度的Seeds和Wine数据集上,与次优方法LDA-NgL1（non-greedy L1-norm linear discriminant analysis））相比,MFA-L2/L1的准确率高出9%;2）在不同维度的AR和FEI人脸数据集上的实验,验证了模型采用L1损失、采用L2损失提升了模型的判别性;3）在Senthil、Yale、ORL、Caltech和UMIST人脸数据集的仿真实验中,MFA-L2/L1算法呈现出较强鲁棒性,性能排名第1。结论 本文提出了一种基于L2/L1损失的鲁棒性度量学习模型,并推导了一种便捷有效的非贪婪式求解算法,进行了算法收敛性的理论分析。在不同数据集的不同噪声情况下的实验结果表明,所提算法具有较好的识别率和鲁棒性。     

一种新的L1度量Fisher线性判别分析研究

Fisher线性判别分析（Fisher Linear Discriminant Analysis,FLDA）是一种典型的监督型特征提取方法,旨在最大化Fisher准则,寻求最优投影矩阵。在标准Fisher准则中,涉及到的度量为[L2]范数度量,此度量通常缺乏鲁棒性,对异常值点较敏感。为提高鲁棒性,引入了一种基于[L1]范数度量的FLDA及其优化求解算法。实验结果表明：在很多情形下,相比于传统的[L2]范数FLDA,[L1]范数FLDA具有更好的分类精度和鲁棒性。