因素空间理论下基点分类算法研究

目前，基于因素空间理论的背景基提取算法计算过程复杂，初始化必须依赖各因素极值，基点数量提取冗余等原因，未能在应用中取得很好效果。为此，结合内点判别法和知识可继承、可扩展的思想，提出一种计算简单、初始化独立、基点数量小的改进的背景基提取算法。然后，利用改进的背景基提取算法构造出一种全新的数据分类算法-基点分类算法，基点分类算法以提取每一类样本的背景基为预测模型，再通过新定义的λ-背景基，优化预测模型。数值实验表明:基点分类算法原理简单、构造难度小、分类模型泛化能力强，预测能力准确率高，同时严格的模型限定区域又能为识别新类别提供新方法。     

Analysis of thin walled structures using the boundary element method

In this article the direct boundary element formulations of stretching and bending plate problems are combined together to solve thin walled structures and other similar frames. The final system is obtained by assuming each individual plane structural element as a subregion. After the necessary transforms of these equations they can be combined together taking into account the displacement compatibility and equilibrium conditions.     

A basis for the design of a multiattribute heuristic for single resource project scheduling

Two important characteristics of a project network are the network's topology and the amount of resources available. Most published project scheduling procedures take one or both of these two characteristics into account. Project scheduling procedures that are robust over variations of network characteristics are desirable. The degree of robustness is generally gauged by one or more measures of performance.

The objective of this paper is to compare a set of priority rules that are useful for a single resource project scheduling problem. The set of networks suggested by Patterson (1973) is used along with the set of performance measures proposed by Khattab and Choobineh 1991. The result of this comparison allows one to design a multiattribute heuristic for a single resource project scheduling problem.     

间作下根际微生物控制土传病害的研究进展

为了解根际微生物在间作模式下有效控制土传病害的作用和机理,综述了间作系统对于根际微生物群落结构和代谢功能的影响,介绍了间作作物组合、作物品种选择和间作方式对根际微生物群落的影响,并从养分和宿主感染位点竞争、生防微生物的拮抗作用、植物化感自毒作用的缓解以及植物诱导系统抗性四个方面总结了与根际微生物相关的间作抗病机理。最后,讨论了现阶段间作抗病研究的局限性并对今后的研究方向进行了展望。     

Lanthanum-Magnesium and Lanthanum-Manganese Donor-Acceptor-Codoped Semiconducting Barium Titanate

BaTiO₃ powder doped with La donor and codoped with Mn or Mg acceptor was sintered at 1350°C/1 h in air. For Ladoped BaTiO₃, the room-temperature resistivity decreased to a minimum at [La³⁺] ∼ 0.15 mol%. For La-Mn-codoped BaTiO₃, the minimum resistivity occurred at [La³⁺] - 2[Mn²⁺] ∼ 0.15 mol%. When the ceramic was changed to a fine-grained insulator by high donor doping ([La³⁺] >0.15 mol%), its semiconductivity was restored, and the relatively homogeneous, coarse-grained microstructure recurred by codoping with either Mg or Mn acceptor, with the transition at [La³⁺] - 2[Mg²⁺] = 0.15 mol% or [La³⁺] - 2[Mn²⁺] = 0.15 mol%. The analogy of a compensation effect between La-Mn- and La-Mg-codoped BaTiO₃ suggested that Mn acceptor added to BaTiO₃ exists as Mn²⁺ ion in the bulk grain region; its influence on the positive temperature coefficient of resistivity behavior is then discussed.     

Nanoindentation Characterization of Submicro- and Nano-Sized Liquid-Phase-Sintered SiC Ceramics

Submicro- and nano-sized liquid-phase-sintered SiC ceramics were mechanically tested by nanoindentation in the peak load range 5–400 mN. The submicro-sized sample showed a marked indentation size effect which the nano-sized samples did not exhibit. The relevance of indentation depth with respect to the microstructural scale has been outlined. In the investigated grain-size range, the hardness dependence on the grain size could be described by a load-dependent inverse Hall–Petch relation. Young's modulus was less microstructure- and load-dependent. Because of the very fine microstructure, the nano-sized SiC materials gave lower elastic values than the submicro-sized SiC ceramic.     

融入类贡献抑制因子的灰度级模糊C均值图像分割

基于灰度级模糊C均值图像分割算法具有分割速度快的优势。由于无损检测图像中背景类和目标类差异较大,该算法不能有效地将目标分割出来,故提出改进的基于灰度级的模糊C均值算法。构建了一种与类大小反向相关的类贡献抑制因子表达式,将之融入目标函数后能够降低较大类对目标函数的贡献,这可避免较小类的聚类中心受较大类的影响而靠近较大类的聚类中心。最小化新的目标函数可得新形式的隶属度和聚类中心表征形式。采用类大小差异较大的无损检测图像进行试验,结果显示本文算法得到的分割图像视觉效果良好,而且指标G_mean也更高,进一步提升了基于灰度级模糊C均值算法适应能力。     

基于智能计算的脑机制研究

解析脑认知原理和脑疾病的发病机制已是当下脑科学研究的热点。智能计算在脑影像领域的应用可为认知心理学研究、脑疾病的识别和干预以及类脑智能理论研究等提供借鉴和参考。本文介绍利用智能影像计算方法在识别脑认知网络成分、构建脑有向连接模型、预测认知行为变化、构建心理过程神经表征等方面的工作,并对未来相关研究进行展望。     

基于图嵌入的自适应多视降维方法

随着监控摄像头的普及和数据采集技术的快速发展,多视数据呈现出规模大、维度高和多源异构的特点,使得数据存储空间大、传输慢、算法复杂度高,造成“有数据、难利用”的困境。到目前为止,国内外在多视降维方面的研究还比较少。针对这一问题,本文提出一种基于图嵌入的自适应多视降维方法。该方法在考虑视角内降维后数据重构原始高维数据的基础上,提出自适应学习相似矩阵来探索不同视角之间降维后数据的关联关系,学习各视数据的正交投影矩阵实现多视降维任务。本文在多个数据集上对降维后的多视数据进行了聚类/识别实验验证,实验结果表明基于图嵌入的自适应多视降维方法优于其他降维方法。     

对抗样本三元组约束的度量学习算法

针对已有三元组约束的度量学习算法大多利用先验知识构建约束,一定程度上制约了度量学习算法性能的问题,本文借鉴对抗训练中样本扰动的思想,在原始样本附近学习对抗样本以构造对抗三元组约束,基于对抗三元组和原始三元组约束构建了度量学习模型,提出了对抗样本三元组约束的度量学习算法(metric learning algorithm with adversarial sample triples constraints,ASTCML)。实验结果表明,提出的算法既克服了已有固定约束方法受先验知识影响大的问题,也提高了分类精度,说明区分更加难以区分的三元组约束能够提升算法的性能。