一种多模态知识图谱实体对齐方法

多模态知识图谱的融合需要解决知识融合过程中的实体对齐问题。在多模态知识图谱中,多模态属性可以提供关键对齐信息来提升实体对齐的能力。本文提出一种基于多模态属性嵌入和图注意力网络的多模态知识图谱实体对齐方法。首先,根据多模态知识图谱中图像、文本和图谱结构信息,将多模态知识图谱划分成子图;其次,利用图注意力网络提取文本和图结构信息,利用视觉几何组(visual geometry group, VGG)网络提取图像特征信息;然后,将文本、图像和图结构特征生成嵌入表示到向量空间;最后,综合子图的多模态特征和图结构特征用于对齐。实验结果表明,在对齐任务中该模型相比于4种基线模型性能有明显提升(Hits@1、Hits@10和MRR提升了10.64%、5.60%和0.227)。

结合多尺度特征与混淆学习的跨模态行人重识别

跨模态行人重识别研究的重难点主要来自于行人图像之间巨大的模态差异和模态内差异。针对这些问题,提出一种结合多尺度特征与混淆学习的网络结构。为实现高效的特征提取、缩小模态内差异,将网络设计为多尺度特征互补的形式,分别学习行人的局部细化特征与全局粗糙特征,从细粒度和粗粒度两方面来增强网络的特征表达能力。利用混淆学习策略,模糊网络的模态识别反馈,挖掘稳定且有效的模态无关属性应对模态差异,来提高特征对模态变化的鲁棒性。在大规模数据集SYSU-MM01的全搜索模式下该算法首位击中率和平均精度(mean average precision,mAP)的结果分别为76.69%和72.45%,在RegDB数据集的可见光到红外模式下该算法首位击中率和mAP的结果分别为94.62%和94.60%,优于现有的主要方法,验证了所提方法的有效性。

分体空调器室外机钣金件动态特性分析及优化设计

本文针对某分体空调室外机低频异响问题,通过现代声振测试手段确定了钣金件与异响之间存在的关联,重点以钣金件为研究对象,通过对钣金件模态属性的测试,结合有限元分析,掌握了钣金件的动态特性,明确了其存在的问题缺陷,在不增加钣金件成本的前提下通过钣金件的局部结构修改对其模态属性等动态特性进行优化设计,修改后样机实验证明本文采取的结构修改措施可以显著改善钣金件声辐射性能,消除室外机低频异响。

使用MFC的UG二次开发数控后置处理器的研究

首先介绍了UG二次开发工具及其工作原理;通过基于UG的二次开发数控后置处理器的研究,给出了在VC++编译环境下使用MFC开发UG应用程序的方法、步骤及界面实现技术。研究结果表明使用MFC开发UG应用程序,可以充分利用VisualC++资源,方便了开发人员的编程工作,缩短了UG二次开发的时间。

Seismic attribute extraction based on HHT and its application in a marine carbonate area

The Hilbert-Huang transform（HHT） is a new analysis method suitable for nonlinear and non-stationary signals.It is very appropriate to seismic signals because they show both non-stationary and nonlinear characteristics.We first introduce the realization of HHT empirical mode decomposition（EMD） and then comparatively analyze three instantaneous frequency algorithms based on intrinsic mode functions（IMF） resulting from EMD,of which one uses the average instantaneous frequency of two sample intervals having higher resolution which can determine that the signal frequency components change with time.The method is used with 3-D poststack migrated seismic data of marine carbonate strata in southern China to effectively extract the three instantaneous attributes.The instantaneous phase attributes of the second intrinsic mode functions（IMF2） better describe the reef facies of the platform margin and the IMF2 instantaneous frequency attribute has better zoning.Combining analysis of the three IMF2 instantaneous seismic attributes and drilling data can identify the distribution of sedimentary facies well.

Adaptive associative classification with emerging frequent patterns

In this paper, we propose an enhanced associative classification method by integrating the dynamic property in the process of associative classification. In the proposed method, we employ a support vector machine（SVM） based method to refine the discovered emerging ~equent patterns for classification rule extension for class label prediction. The empirical study shows that our method can be used to classify increasing resources efficiently and effectively.

Description of the properties of the low-lying energy states in ^(100)Mo with IBM2

The properties of the low-lying energy states for the l00 Mo isotope is investigated within the framework of the proton-neutron interacting model IBM2. By considering the relative energy of the d proton boson to be different from that of the neutron boson and taking into account the dipole interacting among like-boson Lπ·Lπ and Lπ·Lπ, the low-lying energy spectrum is reproduced well. Particularly, the relative position of the energies for 2＋1, 0＋2, 2＋2 and 4＋1 states shifted correctly fit the experimental data. The electromagnetic properties, including the key observable B（E2） reduced transition branching ratios and the E2 reduced matrix elements of the experimental data, are well described. Our calculations show possible shape coexistence in the l00Mo nucleus.