Contrastive Consistency and Attentive Complementarity for Deep Multi-View Subspace Clustering

Deep multi-view subspace clustering (DMVSC) based on self-expression has attracted increasing attention dueto its outstanding performance and nonlinear application. However, most existing methods neglect that viewprivatemeaningless information or noise may interfere with the learning of self-expression, which may lead to thedegeneration of clustering performance. In this paper, we propose a novel framework of Contrastive Consistencyand Attentive Complementarity (CCAC) for DMVsSC. CCAC aligns all the self-expressions of multiple viewsand fuses them based on their discrimination, so that it can effectively explore consistent and complementaryinformation for achieving precise clustering. Specifically, the view-specific self-expression is learned by a selfexpressionlayer embedded into the auto-encoder network for each view. To guarantee consistency across views andreduce the effect of view-private information or noise, we align all the view-specific self-expressions by contrastivelearning. The aligned self-expressions are assigned adaptive weights by channel attention mechanism according totheir discrimination. Then they are fused by convolution kernel to obtain consensus self-expression withmaximumcomplementarity ofmultiple views. Extensive experimental results on four benchmark datasets and one large-scaledataset of the CCAC method outperformother state-of-the-artmethods, demonstrating its clustering effectiveness.

Optimized air-ground data fusion method for mine slope modeling

Refined 3D modeling of mine slopes is pivotal for precise prediction of geological hazards.Aiming at the inadequacy of existing single modeling methods in comprehensively representing the overall and localized characteristics of mining slopes,this study introduces a new method that fuses model data from Unmanned aerial vehicles(UAV)tilt photogrammetry and 3D laser scanning through a data alignment algorithm based on control points.First,the mini batch K-Medoids algorithm is utilized to cluster the point cloud data from ground 3D laser scanning.Then,the elbow rule is applied to determine the optimal cluster number(K0),and the feature points are extracted.Next,the nearest neighbor point algorithm is employed to match the feature points obtained from UAV tilt photogrammetry,and the internal point coordinates are adjusted through the distanceweighted average to construct a 3D model.Finally,by integrating an engineering case study,the K0 value is determined to be 8,with a matching accuracy between the two model datasets ranging from 0.0669 to 1.0373 mm.Therefore,compared with the modeling method utilizing K-medoids clustering algorithm,the new modeling method significantly enhances the computational efficiency,the accuracy of selecting the optimal number of feature points in 3D laser scanning,and the precision of the 3D model derived from UAV tilt photogrammetry.This method provides a research foundation for constructing mine slope model.

Driving fatigue fusion detection based on T-S fuzzy neural network evolved by subtractive clustering and particle swarm optimization

In order to improve the accuracy and reliability of the driving fatigue detection based on a single feature, a new detection algorithm based on multiple features is proposed. Two direct driver＇s facial features reflecting fatigue and one indirect vehicle behavior feature indicating fatigue are considered. Meanwhile, T-S fuzzy neural network（TSFNN）is adopted to recognize the driving fatigue of drivers. For the structure identification of the TSFNN, subtractive clustering（SC） is used to confirm the fuzzy rules and their correlative parameters. Moreover, the particle swarm optimization （PSO）algorithm is improved to train the TSFNN. Simulation results and experiments on vehicles show that the proposed algorithm can effectively improve the convergence speed and the recognition accuracy of the TSFNN, as well as enhance the correct rate of driving fatigue detection.

Nuclear fusion from Coulomb explosions of deuterated methane clusters subjected to ultraintense femtosecond laser pulses

This paper reports that Coulomb explosions taken place in the experiment of heteronuclear deuterated methane clusters （（CD4）n） in a gas jet subjected to intense femtoseeond laser pulses （170 mJ, 70 fs） have led to table-top laser driven DD nuclear fusion. The clusters produced in supersonic expansion had an average size of about 5 nm in radius and the laser intensity used was 3 × 10^17 W/cm^2.The measured maximum and average energies of deuterons produced in the laser-cluster interaction were 60 and 13.5 keV, respectively. Prom DD collisions of energetic deuterons, a yield of 2.5（±0.4） × 10^4 fusion neutrons of 2.45 MeV per shot was realized, giving rise to a neutron production efficiency of about 1.5 × 10^5 per joule of incident laser pulse energy. Theoretical calculations were performed and a fairly good agreement of the calculated neutron yield with that obtained from the present experiment was found.

Effect of laser spot size on fusion neutron yield in laser-deuterium cluster interactions

The effect of the laser spot size on the neutron yield of table-top nuclear fusion from explosions of a femtosecond intense laser pulse heated deuterium clusters is investigated by using a simplified model, in which the cluster size distribution and the energy attenuation of the laser as it propagates through the cluster jet are taken into account. It has been found that there exists a proper laser spot size for the maximum fusion neutron yield for a given laser pulse and a specific deuterium gas cluster jet. The proper spot size, which is dependent on the laser parameters and the cluster jet parameters, has been calculated and compared with the available experimental data. A reasonable agreement between the calculated results and the published experimental results is found.

Evolution of catalytic activity driven by structural fusion of icosahedral gold cluster cores

Atomically precise gold cluster catalysts have emerged as a new frontier in catalysis science,owing to their unexpected catalytic properties.In this work,we explore the evolution of the catalytic activity of clusters formed by the structural fusion of icosahedral Au13 units,namely Au25(SR)18,Au38(SR)24,and Au25(PPh3)10(SC2H4Ph)5Cl2,in the oxidation of pyrrolidine toγ-butyrolactam.We demonstrate that the structural fusion of icosahedral Au13 units,forming vertex-fused(vf),face-fused(ff),and body-fused(bf)clusters,can induce a decrease in the catalytic activity in the following order:Aubf>Auff>Auvf.The structural fusion of icosahedral Au13 units in the clusters does not distinguish the adsorption modes of pyrrolidine over the three clusters from each other,but modulates the chemical adsorption capacity and electronic properties of the three clusters,which is likely to be the key reason for the observed changes in catalytic reactivity.Our results are expected to be extendable to study and design atomically defined catalysts with elaborate structural patterns,in order to produce desired products.

Application of Clustering Fusion on Foreign Fibers Measure-Control System

According to the structure characteristics of foreign fibers detection system,the foreign fiber flow flux mathematical model and fiber detection system were designed.The information fusion clustering structure of foreign fiber flow flux was put forward.The data of the pressure difference,pressure,temperature,and density sensor which had impacted on flux were integrated and output by the Adaptive Resonance Theory-2(ART-2)network and BP network to clustering analysis of output space.The clustering control strategy will keep the output flow pressure stable,when the output pressure and temperature change.

Algorithm for Multi-laser-target Tracking Based on Clustering Fusion

Multi-laser-target tracking is an important subject in the field of signal processing of laser warners. A clustering method is applied to the measurement of laser warner, and the space-time fusion for measurements in the same cluster is accomplished. Real-time tracking of multi-laser-target and real-time picking of multi-laser-signal are introduced using data fusion of the measurements. A prototype device of the algorithm is built up. The results of experiments show that the algorithm is very effective.

Deuterium Clusters Fusion Induced by the Intense Femtosecond Laser Pulse

Neutrons （2.45MeV） from deuterium cluster fusion induced by the intense femtosecond （3Ors） laser pulse are experimentally demonstrated. The average neutron yield 103 per shot is obtained. It is found that the yield slightly increases with the increasing laser spot size. No neutron can be observed when the laser intensity I 〈 4.3 × 10^15 W/cm^2.

基于融合矩阵的文本相似度计算实现检索结果聚类

目的/意义弥补医学文本语义表示方面的不足,实现PubMed数据库检索结果聚类。方法/过程采用Jaccard系数和TF-IDF构建融合矩阵方法,建立短语间、文档间、短语与文档内容间的相似性关系融合矩阵,训练聚类算法,将PubMed数据库检索结果集合分组,随后生成类别标签,描述每一类簇文档的含义。结果/结论基于融合矩阵的聚类效果较好,提取出描述类别的高频词能很好地区分类别含义,对检索结果文本聚类任务有效。

基于改进YOLOv网络的外观检测研究

外观检测涉及对图像或视频中的物体进行准确和高效的识别和定位,为了解决物体表面小尺寸目标检测的问题,研究通过优化YOLOv3网络模型,引入多尺度检测和深度可分离卷积技术来提高检测精度和模型效率,以增强对小尺寸目标的识别能力,再采用深度可分离卷积技术来减少计算量,并提高模型的训练效果;实验结果表明,研究模型在物体表面小尺寸检测方面取得显著提升;与其他金属表面损伤检测算法相比,优化后的YOLOv3实现了71.52%的检测精度,超越Faster R-CNN 6.83%;尽管Faster R-CNN在准确性方面优异但速度慢,SSD速度较快但不及YOLOv2;而YOLOv2虽速度快但精度稍低;相对于原始模型,研究算法的平均精度提升了7.77个百分点,达到了79.21%;虽然网络深度的提升稍增计算量,略有检测速率下降,但引入深度可分离卷积后,检测速度达到36.2帧/秒,仅较原模型稍低2.4帧/秒;研究可以优化算法,提高小尺寸目标检测的准确性和鲁棒性,推动其在计算机视觉领域的广泛应用。

多特征融合及聚类分析的道岔转辙机退化状态识别

针对转辙机退化阶段难以划分的问题,提出一种基于多维特征融合的道岔转辙机退化状态识别方法。首先,提取了S700K转辙机退化功率数据的时域、频域、时频域多域特征;其次,通过核主成分分析(Kernel Principal Components Analysis,KPCA)进行特征融合,获得表征道岔转辙机运行状态的特征向量,构建转辙机退化性能指标;再次,采用K-medoids聚类算法对道岔转辙机性能退化状态进行阶段划分,识别不同的退化状态;最后,选用轮廓系数、分类系数、平均模糊熵对聚类效果进行综合评价,并与模糊C均值聚类(Fuzzy C-Means Clustering,FCM)和古斯塔夫森-凯塞尔(Gustafson Kessel,GK)聚类算法进行比较。研究结果表明,融合特征聚类后的综合评价指标高于单一特征,更能够体现道岔转辙机退化过程中的细节,K-medoids聚类效果明显,模型的准确率达到96.3%,能够对道岔转辙机性能退化阶段进行准确的划分,为铁路现场道岔智能运维提供理论支持。

多示例学习的簇频繁性分析及双角度融合嵌入

多示例学习(Multi-Instance Learning,MIL)的训练数据是由若干个未带标记的示例组成的带标记的包,基于嵌入的方法,通过将包嵌入成单向量来解决包表示问题,然而大部分现有方法忽略了示例与包的联系,难以保证所选示例的代表性.同时,单角度的嵌入方法无法有效地提取正、负包的差异信息,使嵌入向量的质量较差.提出一种多示例学习的簇频繁性分析及双角度融合嵌入(FADE).簇频繁性分析技术从正、负子空间中分别筛选部分示例作为子空间的簇心,依据簇心将子空间聚类成簇,再计算簇频繁性指标,选择频繁性较高的簇的簇心组成子空间代表示例集.双角度融合嵌入技术基于正、负子空间代表示例集和差值嵌入函数,分别从正、负角度挖掘信息,融合两个角度信息获得最终的嵌入向量.在29个数据集上与七个MIL算法进行了对比实验,结果表明,FADE的分类准确率总体上优于七个对比算法,在图像数据集上有显著优势,在文本和网页数据集上也表现良好.

自适应相似图联合优化的多视图聚类

相比于单一视图学习,多视图学习往往可以获得学习对象更全面的信息,因而在无监督学习领域,多视图聚类受到了研究者的极大关注,其中基于图的多视图聚类,近年来取得了很大的研究进展.基于图的多视图聚类一般是先从各个视图原始数据学习相似图,再进行视图间相似图的融合来获得最终聚类结果,因此,多视图聚类的效果是由相似图质量和相似图融合方法共同决定的.然而,现有基于图的多视图聚类方法几乎都聚焦在视图间相似图的融合方法研究上,而缺乏对相似图本身质量的关注.这些方法大多数都是孤立地从各视图的原始数据中学习相似图,并且在后续图融合过程中保持相似图不变.这样得到的相似图不可避免地包含噪声和冗余信息,进而影响后续的图融合和聚类.而少量考虑相似图质量的研究,要么相似图构造和图融合过程是直接联立迭代的,要么在预定义相似图过程中提前利用秩约束进一步初始化,要么就是利用相似图存在的一些底层结构来获取融合图的.这些方法对相似图本身改进很小,最终聚类性能提升也十分有限.同时现有基于图的多视图聚类流程也缺乏对各视图间一致性和不一致性的综合考虑,这也会严重影响最终的多视图聚类性能.为了避免低质量预定义相似图对聚类结果的不利影响以及综合考虑视图间一致性与不一致性来提升最终聚类效果,本文提出了一种自适应相似图联合优化的多视图聚类方法.首先通过Hadamard积来获得视图间高质量一致性部分信息,再将每个预定义相似图和这部分信息对标,重构各个视图的预设相似图.这个过程强化了各视图间的一致性部分,弱化了不一致性部分.其次设计了相似图重构改进和图融合联合迭代优化框架,实现了相似图的自适应改进,最终达到相似图和聚类结果共同提升的效果.该方法将相似图改进过程与图融合过程联合起来进行自适应迭代优化,并且在迭代优化中不断强化各视图间的一致性,弱化视图间的不一致性.此外,本文的方法也集成了现有多视图聚类方法的一些优点,自加权以及无需额外聚类步骤等.在九个基准数据集上与八个对比方法的实验验证了本文方法的有效性与优越性.

基于多特征融合的高机动多目标低截获概率跟踪技术

在多目标跟踪过程中,目标的高机动特性使得传统采用固定运动模型或交互式多模型的目标跟踪算法很难实时精确匹配目标运动模型,从而引起高机动目标的低跟踪精度问题。针对这一问题,本文提出一种基于目标运动状态模型自适应更新的高机动多目标跟踪算法。在多目标跟踪过程中,该算法采用多特征聚类融合算法进行目标运动模型估计,并根据各目标跟踪波动参数进行状态转移矩阵决策更新,同时利用联合概率数据关联实现多机动目标状态转移矩阵自适应更新的关联跟踪,从而解决了传统多目标跟踪算法因目标运动模型失配引起的低跟踪精度问题。在目标跟踪算法的传感器选择上,无源传感器不对外辐射能量,具有较好的低截获概率性能,但其跟踪精度有限,常不能满足多目标高跟踪精度的要求。雷达作为有源传感器,具有较高的跟踪精度。但由于雷达对外辐射信号,容易被防御方截获。针对这一问题,本文提出了一种无源传感器目标跟踪为主,有源雷达间歇跟踪为辅的多传感器协同管理目标跟踪算法。该算法通过对目标跟踪本征堆积误差的判断进行传感器的最优分配,并根据波动参数的大小进行状态转移矩阵决策更新。仿真结果验证了本文所提出的多传感器协同的高机动目标跟踪算法在满足高机动目标跟踪精度的条件下可以有效的提升雷达低截获概率性能。

基于BIRCH聚类的L-Transformer分布式光伏短期发电功率预测

精准的分布式光伏短期发电功率预测有助于电力系统运行与功率就地平衡。该文提出一种基于BIRCH(balanced iterative reducing and clustering using hierarchies)相似日聚类的L-Transformer(LSTM-Transformer)模型进行短期光伏功率预测。首先使用BIRCH无监督聚类算法对历史数据聚类得到3种典型天气,根据聚类结果划分测试集对模型进行训练。为提高不同天气类型下的预测精度,采用双层架构的L-Transformer模型,首层通过长短期记忆网络(long short term memory,LSTM)的门控单元机制捕捉时间序列中的长期依赖关系;次层结合Transformer模型的自注意力机制聚焦于当前任务更关键的特征量,通过多注意力头与光伏数据特征量相结合生成向量,注意力头并行计算,从而高效、精确地预测短期光伏功率。实测数据验证结果表明L-Transformer模型对于不同天气类型功率预测泛化性优异、精确度高,气象数据波动大时鲁棒性强。

靶向药物甲磺酸伊马替尼的分子标志物BCR-ABL融合基因定量检测平台的质量评价方法的研究

目的:使用BCR-ABL定量标准品,评价BCR-ABL融合基因检测试剂盒(数字PCR法)的性能。方法:提取BCR-ABL定量标准品的RNA,测定其浓度和纯度。使用BCR-ABL融合基因定量检测试剂盒(数字PCR法)和数字PCR仪进行检测,得到标准品的BCR-ABL融合基因的分子学反应。结果:用于准确度项目检测的标准品WS2和WS3的BCR-ABL融合基因的MR绝对偏差均不超过±0.5 log,用于检出限项目检测的标准品WS4能检出BCR-ABL融合基因突变阳性,用于重复性项目检测的标准品WS1和WS4的BCR-ABL融合基因的MR的变异系数(CV)均<3.0%。结论:BCR-ABL融合基因定量检测试剂盒(数字PCR法)的准确度、检出限和重复性的性能指标符合制定的《断裂点簇集区-艾贝尔逊白血病病毒(BCR-ABL)融合基因检测试剂盒》标准的相应要求,为标准的实施提供了技术支撑。

基于模糊理论的输电网络电压无功控制策略

为解决新能源、电动汽车、储能等新技术应用下,高复杂度电力系统电压稳定控制问题,提出了一种基于模糊理论的输电网络电压无功控制策略。该方法引入模糊理论中的隶属度函数,根据系统节点与不同分区之间的耦合程度制定无功控制策略。根据灵敏度计算网络各个节点之间的电气距离,通过模糊聚类算法对节点进行初步分区,并采用聚类融合算法对聚类产生的多个结果进行融合,从而得到最终分区结果。根据关键节点对各个分区的隶属度制定主辅控制策略。IEEE30节点输电网络的算例分析表明,该控制策略可以有效实现对无功功率的控制。

针对冲击性故障信号的谱融合特征提取算法

利用盲解卷积方法在时域中进行故障信号特征提取时,常会出现多个信号混淆分离结果,但以往的研究中只强调了分离的部分,而很少对分离后的信号进行进一步的处理,给实际应用造成不便。这里在盲解卷积和谱融合的基础之上,使用核改进的模糊c均值聚类算法,针对机械故障信号的脉冲特性,提出一种针对冲击性故障信号处理的实用型算法。计算机仿真实验证实了该算法的有效性。此算法优化了以往的聚类筛选方法,可以有效排除反卷积后诸多无用信号的干扰,将故障脉冲信号的特征准确提取出来,能提高故障诊断的效率。