L_(2,1)范数正则化的不相关判别分析及其在人脸识别中的应用

对高维数据降维并选取有效特征对分类起着关键作用。针对人脸识别中存在的高维和小样本问题,从特征选取和子空间学习入手,提出了一种L_(2,1)范数正则化的不相关判别分析算法。该算法首先对训练样本矩阵进行奇异值分解;然后通过一系列变换,将原非线性的Fisher鉴别准则函数转化为线性模型;最后加入L_(2,1)范数惩罚项进行求解,得到一组最佳鉴别矢量。将训练样本和测试样本投影到该低维子空间中,利用最近欧氏距离分类器进行分类。由于加入了L_(2,1)范数惩罚项,该算法能使特征选取和子空间学习同时进行,有效改善识别性能。在ORL、YaleB及PIE人脸库上的实验结果表明,算法在有效降维的同时能进一步提高鉴别能力。     

基于属性自表达的低秩超图属性选择算法*

针对高维数据具有低秩形式和属性冗余等特点,提出一种基于属性自表达的无监督超图属性选择算法。具体地,该算法首先利用属性自表达特点用其他属性稀疏地表达每个属性,此自表达形式使用低秩假设寻找高维数据的低秩表示,然后建立超图正则化因子保持高维数据的局部结构,最后利用稀疏正则化因子进行属性选择。属性自表达特性确定属性的重要性,低秩表示相当于考虑数据的全局信息进行子空间学习,超图正则化因子考虑数据的局部结构对数据进行子空间学习。该算法实际上考虑数据全局和局部信息进行子空间学习,更是一种嵌入了子空间学习的属性选择算法。实验结果表明,该算法相比其它对比算法,能更有效地选取属性,并能取得很好的分类效果。     

基于迁移子空间学习的偏最小二乘回归软测量方法

针对流程工业中工况改变易导致当前样本与历史样本分布失配,传统软测量模型失准的问题,考虑工业数据时序性、动态性以及存在过程漂移等特性对建模的影响,提出一种基于迁移子空间学习的偏最小二乘回归软测量方法.首先,回归框架采用非线性迭代偏最小二乘方法,对其求解映射向量的目标函数施加基于子空间重构的域适应正则项,映射过程中保证当前工况中每个样本能够被历史工况样本线性重构.在此基础上对重构矩阵施加低秩稀疏约束,保持数据结构的同时使重构矩阵具备块状结构以应对过程漂移特性.将所提出方法在1个数值案例和3个不同的多工况数据集中进行实验,并与现有域适应回归方法进行对比分析.实验表明,所提出方法能够有效提高模型在跨工况条件下的预测精度,减少工况间数据分布差异对模型性能的影响.     

L_(2,1)范数正则化的广义核判别分析及其人脸识别

特征选取和子空间学习是人脸识别的关键问题。为更准确选取人脸中丰富的非线性特征,并解决小样本问题,提出了一种新的L_(2,1)范数正则化的广义核判别分析(generalized kernel discriminant analysis based on L_(2,1)-norm regularization,L21GKDA)。利用核函数将原始样本隐式地映射到高维特征空间中,得到广义核Fisher鉴别准则,再利用一种有效变换将该非线性模型转化为线性回归模型;为了能使特征选取和子空间学习同时进行,在模型中加入了一种L_(2,1)范数惩罚项,并给出该正则化方法的求解算法。因为方法借助于L_(2,1)范数惩罚项的特征选取能力,所以它能有效地提高识别率。在ORL、AR和PIE人脸库上的实验结果表明,新算法能有效选取人脸的非线性特征,提高判别能力。     

基于稀疏学习的鲁棒自表达属性选择算法

受属性选择处理高维数据表现的高效性和低秩自表达方法在子空间聚类上成功运用的启发,提出一种基于稀疏学习的自表达属性选择算法。算法首先将每个属性用其他属性线性表示得到自表达系数矩阵;然后结合稀疏学习的理论(即整合L2,1-范数为稀疏正则化项惩罚目标函数)实现属性选择。在以分类准确率和方差作为评价指标下,相比其他算法,实验结果表明该算法可更高效地选择出重要属性,且显示出非常好的鲁棒性。     

基于低秩和图拉普拉斯的属性选择算法

针对无监督属性选择算法使用单一方法,未考虑数据间内在相关性和噪声等问题,提出一种基于属性自表达的低秩无监督属性选择算法。算法首先将稀疏正则化（[l2,1-]范数）引入属性自表达损失函数中实现无监督稀疏学习,其次在系数矩阵中加入低秩约束以降低噪声和离群点的影响,然后利用低秩结构和图拉普拉斯正则化使子空间学习兼顾数据的全局和局部结构,最后通过属性自表达实现无监督学习。经数据集上多次迭代验证,该算法能够快速收敛并达到全局最优,与SOGFS、PCA、LPP、RSR等四种算法相比分类准确率平均提高了16.11%、14.03%、9.92%和4.2%,并且在各数据集上互信息平均值也是最高的,说明该算法有效、高效。     

基于跨域流形正则化特征域适应的浮选工况识别

针对浮选过程的故障工况信息不足难以建立准确识别模型,导致调整浮选生产工况不及时,从而无法正常稳定运行的问题,提出一种基于跨域流形正则化特征域适应方法.该方法将已有相似完备浮选过程积累的丰富工况信息作为源域迁移至未建模的不完备浮选过程的目标域中,首先,通过最大域内类密度和局部流形正则化约束分别保留原始判别信息和维持域内邻域结构信息不变,从而提取完备工况与不完备工况域间的特征并投影至公共子空间;然后,由最大均值差异缩小源域与目标域间分布差异,建立分类识别模型,再结合D-S证据理论,融合浮选过程泡沫的静态特征与动态特征信息,提高对不完备浮选过程工况识别的泛化能力,保证得到较好的识别分类效果;最后,通过仿真实验验证所提出方法的有效性.     

引入低秩约束先验的深度子空间聚类

大多数子空间聚类算法将高维数据映射到低维子空间时不能较好捕获数据间几何结构.针对上述问题,文中提出引入低秩约束先验的深度子空间聚类算法,兼顾数据全局和局部结构信息.算法结合低秩表示与深度自编码器,利用低秩约束捕获数据全局结构,并将约束神经网络的潜在特征表示为低秩.自编码通过最小化重构误差进行非线性低维子空间映射,保留数据的局部特性.以多元逻辑回归函数作为判别模型,预测子空间分割.整个算法在无监督联合学习框架下进行优化.在5个数据集上的实验验证文中方法的有效性.     

拉普拉斯正则化双曲正切低秩子空间聚类算法

传统的低秩表示模型LRR对高维数据聚类精确度低,针对这一情况提出一种基于拉普拉斯正则化双曲正切函数低秩子空间聚类算法(LRHT-LRSC).该算法利用双曲正切函数代替核范数以便更紧凑地逼近秩函数,并利用拉普拉斯正则项刻画数据本身的几何结构,提高了数据聚类的准确率;然后构建数据样本的系数矩阵和相似矩阵;最后利用谱聚类方法得到最终的聚类结果.在合成数据集、真实数据集ExtendedYaleB和Hopkins155上的对比实验结果表明,LRHT-LRSC能够提高聚类的准确率和鲁棒性.     

基于自监督知识的无监督新集域适应学习

无监督域适应(unsupervised domain adaptation,UDA)旨在利用带大量标注数据的源域帮助无任何标注信息的目标域学习.在UDA中,通常假设源域和目标域间的数据分布不同,但共享相同的类标签空间.但在真实开放学习场景中,域间的标签空间很可能存在差异.在极端情形下,域间的类别不存在交集,即目标域中类别都为新未知类别.此时若直接迁移源域的类判别知识,可能会损害目标域性能,导致负迁移问题.为此,提出了基于自监督知识的无监督新集域适应(unsupervised new-set domain adaptation with self-supervised knowledge, SUNDA)方法,迁移源域的样本对比知识;同时,利用目标域的自监督知识指导知识迁移.首先,通过自监督学习源域和目标域初始特征,并固定部分网络参数用于保存目标域信息.再将源域的样本对比知识迁移至目标域,辅助目标域学习类判别特征.此外,利用基于图的自监督分类损失,解决域间无共享类别时目标域的分类问题.在手写体数字的无共享类别跨域迁移和人脸数据的无共享类别跨种族迁移任务上对SUNDA进行评估,实验结果表明,...     

基于核函数的稀疏属性选择算法

鉴于传统属性选择算法无法捕捉属性之间的关系的问题,文中提出了一种非线性属性选择方法。该方法通过引入核函数,将原始数据集投影到高维的核空间,因在核空间内进行运算,进而可以考虑到数据属性之间的关系。由于核函数自身的优越性,即使数据通过高斯核投影到无穷维的空间中,计算复杂度亦可以控制得较小。在正则化因子的限制上,使用两种范数进行双重约束,不仅提高了算法的准确率,而且使得算法实验结果的方差仅为0.74,远小于其他同类对比算法,且算法更加稳定。在8个常用的数据集上将所提算法与6个同类算法进行比较,并用SVM分类器来测试分类准确率,最终该算法得到最少1.84%,最高3.27%,平均2.75%的提升。     

一种改进的小波域图像修复算法

针对全变分小波图像修复算法在平滑区易产生阶梯效应并且噪声抑制不充分的缺陷,提出一种新颖的小波域图像修复模型。通过优化一个全变分和图像梯度的二范数的能量泛函来实现小波域的图像修复,使得在去除噪声的同时较好地保留了图像的边缘,在平滑区域削弱了阶梯效应,并利用有限差分法对所建立的扩散方程进行数值求解。实验结果表明:所提方法对于不同丢失比例的图像以及噪声图像都取得了较好的修复效果,尤其当小波系数丢失率较高时更为明显。     

基于稀疏特性的盲二值图像去模糊

通过分析二值图像发现其像素值具有稀疏特性,因此采用L₀梯度反卷积算法结合二值图像的组合特性来处理盲二值图像的复原问题。常见的图像复原方法均将二值图像看作灰度值图像来处理,当其考虑到二值图像的特殊性质时,将会针对这种特定类型的图像得到更好的复原效果。提出的盲复原算法基于一阶梯度空间L₀最小化问题的框架,利用L₀梯度图像平滑方法来获得明显的图像边缘以估计模糊核,并将二值图像的特有属性作为正则项加入目标函数。在图像的复原过程中,通过二值图像先验来强制复原结果趋于二值图像。根据提出的模型,给出了基于稀疏特性的盲二值图像复原算法。通过实验将该算法与传统的盲反卷积复原算法进行比较,结果表明所提算法具有良好的性能,对二值图像进行复原是有效的。     

锂-亚硫酰氯电池储存寿命研究

锂-亚硫酰氯电池作为一种免维护、高比能、长储存寿命电池,目前已经在以国防领域为代表的国民经济中得到了广泛应用;其储存寿命的考核在行业内尚属难题;通过广泛、深入地调研和对前期锂-亚硫酰氯电池储存数据的收集整理,研究了锂-亚硫酰氯电池的储存寿命影响因素及其试验评估方法;通过研究得知,锂-亚硫酰氯电池的储存寿命试验应尽早备样,若时间紧迫可通过加速试验方法;提出了通过等效储存试验时间来评估电池储存寿命及其可靠度的方法,指出当等效储存试验时间不足时,应安排样本进行容量回归分析,得出其退化规律;此外,还要对电池储存末期热性能进行分析;在以上工作基础上对电池储存寿命进行综合评估;最后,通过案例分析,进行了工程演算;为后续锂-亚硫酰氯电池储存寿命评估提供了参考。     

基于L2,1范数稀疏特征选择和超法向量的深度图像序列行为识别

结合L_2,1范数稀疏特征选择和超法向量提出了一种新的深度图像序列行为识别方法。首先从深度图像序列中提取超法向量特征;然后利用L_2,1范数稀疏特征选择方法从超法向量特征中选择出最具判别性的稀疏特征子集作为特征表示;最后利用线性分类器Liblinear进行分类。在MSR Action3D数据库上的实验结果表明,所提方法使用2%的超法向量特征获得的识别率为94.55%,并且 具有比 其他方法更高的识别精度。     

土壤水分传感器实验室标定和测试方法探讨

通过在实验室制作标准试样(土壤)标定及测试土壤水分传感器,研究土壤水分传感器在实验室标定及测试的步骤和方法,探讨土壤水分传感器实验室环境下率定关系式的建立,提高土壤水分传感器在实验室环境下数据测试的准确性和权威性,在于探讨土壤水分传感器实验室标定和测试方法,为土壤水分传感器的实验室检测提供方法支撑。     

Simultaneous measurements of plant structure and chlorophyll content in broadleaf saplings with a terrestrial laser scanner

Plant structure and chlorophyll content strongly affect rates of photosynthesis. Rapid, objective, and repeatable methods are needed to measure these vegetative parameters to advance our understanding and modeling of plant ecophysiological processes. Terrestrial laser scanners (TLS) can be used to measure structural and potentially chemical properties of objects by quantifying the x,y,z coordinates and intensity of laser light, respectively, returned from an object's surface. The objective of this study was to determine the potential usefulness of TLS with a green (532 nm) laser to simultaneously measure the spatial distribution of chlorophyll a and b content (Chl_ab), leaf area (LA), and leaf angle (LAN). The TLS measurements were obtained from saplings of two tree species (Quercus macrocarpa and Acer saccharum) and from an angle-adjustable cardboard surface. The green laser return intensity value was strongly correlated with wet-chemically determined Chl_ab (r² = 0.77). Strong agreement was shown between measured and TLS-derived LA (r² = 0.95, intercept = − 1.43, slope = 0.97). The TLS derived LANs of both species followed a plagiophile LAN distribution, and the measured angles of the cardboard surface allowed us to quantify that these LAN values were strongly correlated with TLS derived angles (r² = 1.0, intercept and slope = 0.98). Our results show that terrestrial laser scanners are feasible for simultaneous measurement of LA, LAN, and Chl_ab in simple canopies of small broadleaved plants. Further research is needed in more complex and larger canopies.     

A new instrument for passive remote sensing: 2. Measurement of leaf and canopy reflectance changes at 531 nm and their relationship with photosynthesis and chlorophyll fluorescence

A previously described passive remote sensing fluorimeter (see companion paper) was modified to detect changes in the reflectance of vegetation. The utility of this remote sensing technique to measure the Physiological Reflectance Index (PRI) is shown at both leaf level under laboratory conditions and at the canopy level in the field. PRI, defined as the relative changes in reflectance at 531 nm with respect to those at 570 nm (PRI=R531−R570/R531+R570), is related to xanthophyll-related, dynamic changes of non-photochemical quenching of chlorophyll fluorescence. The robustness of this relationship by simultaneous remote sensing of PRI and chlorophyll fluorescence is strengthened. At the leaf level, the existence of two kinetically distinct components of PRI is shown. A fast (within seconds) component that is partly attributed to ΔpH induced chloroplast shrinkage, and a slow (within minutes), main component that is related to xanthophyll de-epoxidation, as demonstrated by its disappearance in the presence of DTT. Overall, PRI correlated better with non-photochemical quenching of chlorophyll fluorescence (NPQ) than with any other measured parameter, including the photochemical efficiency of PSII. Finally, at the canopy level and under field conditions, it is shown that PRI can be a useful tool for remote sensing of water stress in grapevines.     

Modelling of diesel engine performance using advanced machine learning methods under scarce and exponential data set

Traditional methods on creating diesel engine models include the analytical methods like multi-zone models and the intelligent based models like artificial neural network (ANN) based models. However, those analytical models require excessive assumptions while those ANN models have many drawbacks such as the tendency to overfitting and the difficulties to determine the optimal network structure. In this paper, several emerging advanced machine learning techniques, including least squares support vector machine (LS-SVM), relevance vector machine (RVM), basic extreme learning machine (ELM) and kernel based ELM, are newly applied to the modelling of diesel engine performance. Experiments were carried out to collect sample data for model training and verification. Limited by the experiment conditions, only 24 sample data sets were acquired, resulting in data scarcity. Six-fold cross-validation is therefore adopted to address this issue. Some of the sample data are also found to suffer from the problem of data exponentiality, where the engine performance output grows up exponentially along the engine speed and engine torque. This seriously deteriorates the prediction accuracy. Thus, logarithmic transformation of dependent variables is utilized to pre-process the data. Besides, a hybrid of leave-one-out cross-validation and Bayesian inference is, for the first time, proposed for the selection of hyperparameters of kernel based ELM. A comparison among the advanced machine learning techniques, along with two traditional types of ANN models, namely back propagation neural network (BPNN) and radial basis function neural network (RBFNN), is conducted. The model evaluation is made based on the time complexity, space complexity, and prediction accuracy. The evaluation results show that kernel based ELM with the logarithmic transformation and hybrid inference is far better than basic ELM, LS-SVM, RVM, BPNN and RBFNN, in terms of prediction accuracy and training time.     

A label-free amperometric immunosensor based on horseradish peroxidase functionalized carbon nanotubes and bilayer gold nanoparticles

In this work, a novel label-free amperometric immunosensor has been constructed for detecting α-1-fetoprotein (AFP) based on nanocomposite of horseradish peroxidase (HRP) labeled carbon nanotubes (CNTs). First, the gold nanoparticles (AuNPs) were electrodeposited on the surface of the glass carbon electrode by electrochemical reduction of gold chloride tetrahydrate (HAuCl₄) to immobilize horseradish peroxidase labeled carbon nanotubes (HRP-CNTs). Then HRP-CNTs bioconjugate was immobilized on the surface of the electrodeposited AuNPs layer by the combination of forces (coordination and electrostatic force). Subsequently, it was immersed into gold colloidal nanoparticles (GNPs) solution, which was used to immobilize antibody biomolecules (anti-AFP). Enhanced sensitivity was obtained by using bioconjugates featuring HRP labeled (HRP-CNTs), which had lager specific surface area and good electronic catalysis (current response signal) compared to carbon nanotubes. Under optimized conditions, the linear ranges were from 0.2 to 200 ng mL⁻¹ with a detection limit of 0.067 ng mL⁻¹ (at an S/N of 3). The proposed immunosenor showed good precision, acceptable stability and reproducibility and could be used for the detection AFP in normal human serum, which provided a potential alternative tool for the detection of protein in clinical diagnosis.