基于改进网格运动统计特征的图像匹配算法

为了解决尺度不变特征变换（SIFT）算法在图像匹配中匹配正确率低、耗时长等问题，提出一种基于改进网格运动统计特征RANSAC-GMS的图像匹配算法。首先，利用快速旋转不变性特征（ORB）算法对图像进行预匹配，对预匹配的特征点采用网格运动统计（GMS）来支持估计量以实现正确匹配点与错误匹配点的区分；然后，采用改进的随机抽样一致性（RANSAC）算法通过匹配点间的距离相似性对特征点进行筛选，并采用评价函数对筛选后的新数据集进行重新整理，进而实现对误匹配点的剔除。采用Oxford标准图库和现实中拍摄的图像对图像匹配算法进行测试对比，实验结果表明，所提算法在图像匹配中的平均匹配正确率达到91%以上；与GMS、SIFT、ORB等算法相比，该改进算法的近景匹配正确率和远景匹配正确率分别最少提高了16.15个百分点和3.56个百分点，说明它能有效剔除误匹配点，进一步提高图像匹配精度。     

基于间隔理论的过采样集成算法

针对传统集成算法不适用于不平衡数据分类的问题，提出基于间隔理论的AdaBoost算法（MOSBoost）。首先通过预训练得到原始样本的间隔；然后依据间隔排序对少类样本进行启发式复制，从而形成新的平衡样本集；最后将平衡样本集输入AdaBoost算法进行训练以得到最终集成分类器。在UCI数据集上进行测试实验，利用F-measure和G-mean两个准则对MOSBoost、AdaBoost、随机过采样AdaBoost（ROSBoost）和随机降采样AdaBoost（RDSBoost）四种算法进行评价。实验结果表明，MOSBoost算法分类性能优于其他三种算法，其中，相对于AdaBoost算法，MOSBoost算法在F-measure和G-mean准则下分别提升了8.4%和6.2%。     

A globally exponentially stable speed observer for a class of mechanical systems: experimental and simulation comparison with high-gain and sliding mode designs

It is shown in the paper that the problem of speed observation for mechanical systems that are partially linearisable via coordinate changes admits a very simple and robust (exponentially stable) solution with a Luenberger-like observer. This result should be contrasted with the very complicated observers based on immersion and invariance reported in the literature. A second contribution of the paper is to compare, via realistic simulations and highly detailed experiments, the performance of the proposed observer with well-known high-gain and sliding mode observers. In particular, to show that – due to their high sensitivity to noise, that is unavoidable in mechanical systems applications – the performance of the two latter designs is well below par.     

The digital public humanities: giving new arguments and new ways to argue

ABSTRACT

In response to the latest “crisis” in the humanities, advocates have marched, rallied, fundraised, and—especially—argued. This essay contends that communication scholars can support the growing “case for the humanities” by analyzing argumentative strategies, and more specifically, by offering ethical argumentative strategies that avoid replicating structures of domination. In particular, we look to Mari Lee Mifsud's theorization of rhetoric as gift, which follows Henry W. Johnstone in conceptualizing argument as something other than winning over an adversary. We place Mifsud's theorization of the gift in conversation with the methods of the digital public humanities (DPH), which acknowledge and offer abundant resources for meaning-making. Through the methods of DPH, we offer a response to the humanities “crisis” that activates the humanities’ already broad constituencies by giving resources for humanistic inquiry rather than seeking to capture adversaries. Our case study is Photogrammar, a DPH project for organizing, searching, and visualizing the New Deal and World War II era photographs funded by the U.S. federal government. The project forefronts visual, nonlinear, and interactive argumentation in order to engage publics in generative humanistic inquiry. By enlisting participants and sharing expertise, Photogrammar shows how humanities advocates can deepen attachments to the humanities and build broad constituencies of collaborators and allies.     

面向智能客服的句子相似度计算方法

针对金融领域中智能客服的句子相似度计算方法进行了研究。利用基于词性的分词纠正模型减少中文歧义词、金融相关词汇的分词错误；通过词向量方法和循环神经网络分别提取词语级和句子级的语义特征，并且得到句子向量；用融合层计算出句子向量间的差异特征；对差异特征进行降维和归一化得到句子相似度计算结果。实验结果表明，该方法具有较高的准确率和[F1]值。     

基于孪生神经网络的时间序列相似性度量

在时间序列分类等数据挖掘工作中，不同数据集基于类别的相似性表现有明显不同，因此一个合理有效的相似性度量对数据挖掘非常关键。传统的欧氏距离、余弦距离和动态时间弯曲等方法仅针对数据自身进行相似度公式计算，忽略了不同数据集所包含的知识标注对于相似性度量的影响。为了解决这一问题，提出基于孪生神经网络（SNN）的时间序列相似性度量学习方法。该方法从样例标签的监督信息中学习数据之间的邻域关系，建立时间序列之间的高效距离度量。在UCR提供的时间序列数据集上进行的相似性度量和验证性分类实验的结果表明，与ED/DTW-1NN相比SNN在分类质量总体上有明显的提升。虽然基于动态时间弯曲（DTW）的1近邻（1NN）分类方法在部分数据上表现优于基于SNN的1NN分类方法，但在分类过程的相似度计算复杂度和速度上SNN优于DTW。可见所提方法能明显提高分类数据集相似性的度量效率，在高维、复杂的时间序列的数据分类上有不错的表现。     

Enhancing optical absorption in visible light of ZnO co-doped with europium and promethium by first-principles study through modified Becke and Johnson potential scheme

By using first-principles calculations we studied the electronic, optical and magnetic properties of ZnO co-doped with Eu and Pm. In this calculation, we used Wien2k code based on full potential linearized augmented plane waves (FP-LAPW) method with the modified Becke-Johnson (mBJ) approximation. This correction gives good band gap compared to experimental band gap. The introduction of Eu and Pm codoping leads to an increase in the band gap. Electrons can transit easily from the valence band to the conduction band, which results in an enhancement of visible light absorption in a wider absorption range. Absorption spectra reach a high value in visible and infrared light regions. With the significance of the obtained results, the studied compounds may potentially find spintronic and optoelectronic applications.     

Theoretical kinetic study of the reaction between dimethyl disulfide and OH radicals

The potential energy profile of the reaction between dimethyl disulfide and OH^? radicals is explored by utilizing ab initio and hybrid meta density functional theory methods. Having the energies and structural data of the stationary points, statistical rate theories, such as transition state theory and variable reaction coordinate-transition state theory, are employed to compute the overall rate constants, and discuss the mechanism and product channels. On the basis of the calculations, the overall rate coefficient is predicted to be 2.49?×?10^?10?cm³?molecule^?1?s^?1 at 298?K. It is found that in the most favorable pathway, the reaction proceeds via formation of the relatively unstable intermediate CH₃S^?(OH)SCH₃ decomposing rapidly to yield CH₃S^?+CH₃SOH.     

Bond characteristics,vibrational spectrum and optimized microwave dielectric properties of chemically substituted NdNbO4

Coordination chemistry, bond state and vibrational spectrum of co-substituted microwave dielectric NdNb_1-x(Zr_0.5W_0.5)_xO₄ ceramics (x = 0.01∼0.05) were investigated. Raman spectra and XRD refinement showed a solid solution was formed. The compressed and elongated chemical bonds are responsible for the variations of crystal parameters and cell volume. Calculated chemical bond parameters indicated bond covalency, lattice energy and Nb-site bond energy act on the fluctuations of the permittivity, quality factor and temperature coefficient, respectively. Meanwhile, the infrared vibrational spectrum is fitted to quantify the contributions of observed IR mode to the intrinsic loss. Compact ceramic possesses excellent properties: ε_r ∼ 19.2, Q × f ∼ 55282 GHz and τ_f ∼ -11.36 ppm/°C with x = 0.04, at 1250°C.     

基于克里金理论的辐射剂量场插值重构与可视化研究

使用基于克里金理论的插值计算方法对均匀网格与随机排列等不同分布类型的辐射剂量场进行了数据重构与可视化研究,并将重构结果与常用的反距离加权插值法进行了对比,分析了不同插值方法对辐射剂量场重构精度与可视化显示效果的影响。结果表明:对于多数情况,克里金插值法在辐射剂量场重构的精度与可视化显示的效果上均明显优于反距离加权插值法,算例中克里金插值法的结果与原始数据之间的偏差在10%以内,而反距离加权插值法的结果与原始数据之间的偏差基本在15%以上;但在辐射剂量场梯度变化较大且已知数据极少的区域,克里金插值法的计算结果偏差会增大,此时可使用反距离加权插值作为其补充,综合使用。本文的研究工作,验证了克里金插值法在辐射剂量场重构与可视化方面的应用潜力,可为推动辐射剂量场可视化仿真技术的发展提供技术支持。