基于多模态掩码Transformer网络的社会事件分类

多模态社会事件分类的关键是充分且准确地利用图像和文字2种模态的特征。然而,现有的大多数方法存在以下局限性:简单地将事件的图像特征和文本特征连接起来,不同模态之间存在不相关的上下文信息导致相互干扰。因此,仅仅考虑多模态数据模态间的关系是不够的,还要考虑模态之间不相关的上下文信息(即区域或单词)。为克服这些局限性,提出一种新颖的基于多模态掩码Transformer网络(MMTN)模型的社会事件分类方法。通过图-文编码网络来学习文本和图像的更好的表示。将获得的图像和文本表示输入多模态掩码Transformer网络来融合多模态信息,并通过计算多模态信息之间的相似性,对多模态信息的模态间的关系进行建模,掩盖模态之间的不相关上下文。在2个基准数据集上的大量实验表明:所提模型达到了最先进的性能。

Nonlinear amplitude versus angle inversion for transversely isotropic media with vertical symmetry axis using new weak anisotropy approximation equations

In VTI media,the conventional inversion methods based on the existing approximation formulas are difficult to accurately estimate the anisotropic parameters of reservoirs,even more so for unconventional reservoirs with strong seismic anisotropy.Theoretically,the above problems can be solved by utilizing the exact reflection coefficients equations.However,their complicated expression increases the difficulty in calculating the Jacobian matrix when applying them to the Bayesian deterministic inversion.Therefore,the new reduced approximation equations starting from the exact equations are derived here by linearizing the slowness expressions.The relatively simple form and satisfactory calculation accuracy make the reduced equations easy to apply for inversion while ensuring the accuracy of the inversion results.In addition,the blockiness constraint,which follows the differentiable Laplace distribution,is added to the prior model to improve contrasts between layers.Then,the concept of GLI and an iterative reweighted least-squares algorithm is combined to solve the objective function.Lastly,we obtain the iterative solution expression of the elastic parameters and anisotropy parameters and achieve nonlinear AVA inversion based on the reduced equations.The test results of synthetic data and field data show that the proposed method can accurately obtain the VTI parameters from prestack AVA seismic data.

TRANSFORMATION OF EFFUSION AND FORMATION OF GRANULATION TISSUE IN PATHOLOGIC PROCESS OF OTITIS MEDIA WITH EFFUSION

Objective To explore the diversity of stagnant effusions and the pathologic processes leading to granulation tissue formation in otitis media with effusion(OME). Methods Temporal bone slides from 306 ears with OME were studied histopathologically under the light microscope. Results Results of this study revealed a pathologic process in witch the type and the condition of stagnant effusion in the middle ear cleft was a variable in the dynamics of OME progression from an early stage to an advanced stage. The location of granulation tissue and retentive effusion were found to be closely related. Conclusion Early stage granulation tissue formation exhibited a pathologic process in which granulation tissue formation occurred only in areas where effusion had stagnated or was absorbed. The incidence of the retentive effusion and formation of granulation tissue was much higher and the pathologic changes most extensive in the area around the ossicular chain.

Enhancing Arabic Cyberbullying Detection with End-to-End Transformer Model

Cyberbullying,a critical concern for digital safety,necessitates effective linguistic analysis tools that can navigate the complexities of language use in online spaces.To tackle this challenge,our study introduces a new approach employing Bidirectional Encoder Representations from the Transformers(BERT)base model(cased),originally pretrained in English.This model is uniquely adapted to recognize the intricate nuances of Arabic online communication,a key aspect often overlooked in conventional cyberbullying detection methods.Our model is an end-to-end solution that has been fine-tuned on a diverse dataset of Arabic social media(SM)tweets showing a notable increase in detection accuracy and sensitivity compared to existing methods.Experimental results on a diverse Arabic dataset collected from the‘X platform’demonstrate a notable increase in detection accuracy and sensitivity compared to existing methods.E-BERT shows a substantial improvement in performance,evidenced by an accuracy of 98.45%,precision of 99.17%,recall of 99.10%,and an F1 score of 99.14%.The proposed E-BERT not only addresses a critical gap in cyberbullying detection in Arabic online forums but also sets a precedent for applying cross-lingual pretrained models in regional language applications,offering a scalable and effective framework for enhancing online safety across Arabic-speaking communities.

Exact Solution of Second Grade Fluid in a Rotating Frame through Porous Media Using Hodograph Transformation Method

In this paper exact solution for a homogenous incompressible, second grade fluid in a rotating frame through porous media has been provided using hodograph-Legendre transformation method. Results are summarised in the form of theorems. Two examples have been taken and streamline patterns are shown for the solutions.

Recognition Method of Aircraft Axis Direction Based on Morphological Skeleton and Hough Transform

Because of the limit of angle of view(AOV) of IR imaging seeker during the approach of missile and target, the detector can only get the partial image sequence of aircraft nose after "lose point". Recognizing the axis direction on the basis of partial IR image sequence is a key issue of the advanced IR imaging guide air-to-air missile faced. In this paper, a recognition method was proposed based on the morphological skeleton and modified Hough transform, and this method can recognize correctly the axis direction of aircraft nose in different poses during missile-target encounter. Firstly, the morphological skeleton transform was used for the extraction of skeleton features. Secondly, the modified Hough transform was used for the straight-lines detection. Finally, According to the relations between aircraft nose and axis and invariant of nose features in high-speed IR image sequence, the axis direction can be detected and calculated. Experimental results indicate that the method is feasible and effective, and the precision of axis direction recognized can meet the requirement of accurate burst control of GIF fuze.

Large Interactive Media Display and Its Influence on Transformation Urban Spaces from Neglecting to Action: The Case of Al-Thaqafa Street in Amman City

In 21st?century, media become the most important factor affecting the development of urban cities, including public places. As a result of the digital revolution, re-imaging and re-linkage public places by media are essential to create more interactions between public spaces and users, interaction media display, and urban screens, one of the most important defined media. This interaction can transform the urban space from being neglected to be more interactive space with users, specially the pedestrians. This paper aims to identify the effects of these new digital factors to transform public spaces, and the influences of large media display on the interaction between urban spaces and pedestrians. The paper focuses on Al-Thaqafa Street as one of the neglected spaces in Amman city, and attempts to analyse this street, explains its problems, and studies the influence of these new digital factors on its transformation, to be more active and vital by pedestrians.

Symmetric axis detection for images based on Hough algorithm

To extract the symmetric axis o{ rigid target accurately, a symmetric axis detection method is proposed based on Hough algorithm. A bullet is selected as a research object. Firstly, the original image is collected and the characteristics of the target image are analyzed. Because the symmetric axis detection depends on the edge detection of the image, it is necessary to use relevant operators to detect the edge and get all possible edge points. Secondly, all possible symmetric axes related to all contour points acquired are determined by Hough transform, and all possible inclination angles and intercepts and their ranges are obtained. Finally, by using least squares method, when the distance between the symmetric points of the contour points from the one edge and the contour points from the other edge is the minimum, the optimal symmetric axis is got. Simulation resuits show that the proposed method can improve noise-resistance and precision of symmetric axis detection and has certain practical value.

Axis intersection measurement of three-axis turntable with two crosshair targets

In order to measure three-axis intersection error, two crosshair targets were fixed in the inner axis frame of a three-axis turntable. Also a theodolite was used to point its telescope to the targets and to measure the horizontal angles when three axes were on equi-spaced angle positions. The calculation equations of the axis intersection were deduced from the mounting position of the theodolite, positions of two targets, angular positions of three axes, and the measured horizontal angles with the theodolite. Finally, a practical measurement is carried out on a horizontal three-axis turntable and error analysis is conducted.

Automatic Identification of Axis Orbit Based on Both Wavelet Moment Invariants and Neural Network

Axis orbit is an important characteristic to be used in the condition monitoring and diagnosis system of rotating machine. The wavelet moment has the invariant to the translation, scaling and rotation. A method, which uses a neural network based on Radial Basis Function (RBF) and wavelet moment invariants to identify the orbit of shaft centerline of rotating machine is discussed in this paper. The principle and its application procedure of the method are introduced in detail. It gives simulation results of automatic identification for three typical axis orbits. It is proved that the method is effective and practicable.

A Semi-analytical Method for Vibrational and Buckling Analysis of Functionally Graded Nanobeams Considering the Physical Neutral Axis Position

In this paper,a semi-analytical method is presented for free vibration and buckling analysis of functionally graded(FG)size-dependent nanobeams based on the physical neutral axis position.It is the first time that a semi-analytical differential transform method(DTM)solution is developed for the FG nanobeams vibration and buckling analysis.Material properties of FG nanobeam are supposed to vary continuously along the thickness according to the power-law form.The physical neutral axis position for mentioned FG nanobeams is determined.The small scale effect is taken into consideration based on nonlocal elasticity theory of Eringen.The nonlocal equations of motion are derived through Hamilton’s principle and they are solved applying DTM.It is demonstrated that the DTM has high precision and computational efficiency in the vibration analysis of FG nanobeams.The good agreement between the results of this article and those available in literature validated the presented approach.The detailed mathematical derivations are presented and numerical investigations are performed while the emphasis is placed on investigating the effect of the several parameters such as neutral axis position,small scale effects,the material distribution profile,mode number,thickness ratio and boundary conditions on the normalized natural frequencies and dimensionless buckling load of the FG nanobeams in detail.It is explicitly shown that the vibration and buckling behaviour of a FG nanobeams is significantly influenced by these effects.

A Solution of the Burger’s Equation Arising in the Longitudinal Dispersion Phenomenon in Fluid Flow through Porous Media by Mixture of New Integral Transform and Homotopy Perturbation Method

The main aim of the paper is to examine the concentration of the longitudinal dispersion phenomenon arising in fluid flow through porous media. These phenomenon yields a partial differential equation namely Burger’s equation, which is solved by mixture of the new integral transform and the homotopy perturbation method under suitable conditions and the standard assumption. This method provides an analytical approximation in a rapidly convergent sequence with in exclusive manner computed terms. Its rapid convergence shows that the method is trustworthy and introduces a significant improvement in solving nonlinear partial differential equations over existing methods. It is concluded that the behaviour of concentration in longitudinal dispersion phenomenon is decreases as distance x is increasing with fixed time t > 0 and slightly increases with time t.

基于Transformer的小样本细粒度图像分类方法

针对小样本细粒度图像分类任务中存在的相似性度量单一以及细粒度特征提取效果不佳的问题,提出了一种基于Transformer的小样本细粒度图像分类方法,克服了小样本学习在细粒度图像分类中由于样本数量较少从而分类效果较差的问题。构建以多轴注意力模块与卷积算子为基本组件的新模块CBG Transformer Block,通过该模块的重复堆叠提高了网络的特征提取能力;采用关系网络和余弦网络组成的双相似度模块进行相似性度量,避免了在训练数据量较小的情况下单一度量造成的相似性偏差;通过计算两个相似度得分的平均值得出最终预测结果。实验结果表明,提出的方法在CUB-200-2011、Stanford Cars和Stanford Dogs三个公开细粒度图像数据集上的5-way5-shot任务分类精度分别达到了82.70%、74.22%和69.68%,可见在小样本细粒度图像分类任务中取得了优异效果。

Structural transformation of FePt nanocomposite films during annealing and its effects on magnetic properties

Fe100-xPtx(x=30at.%-60at.%) nanocomposite films were deposited on natural-oxidized Si(100) substrates by magnetron sputtering. The as-deposited films were annealed between 373 and 1073 K. In situ X-ray diffraction shows that the FePt nanocomposite films undergo a phase transformation from a disordered FCC phase to an ordered L10 phase between 673 and 773 K. The coercivity is 306 kA·m-1 whiles the average grain sizes is about 10 nm in the optimized FePt alloy film sample annealed at 673 K. The adjustable coercivity and fine grain size suggest that this FePt nanocomposites system is suitable as recording media at extremely high areal density.

A Generic Kinematic Model for Three Main Types of Five-axis Machine Tools

Material removal is one of the most used processes in manufacturing. Five-axis CNC machines are believed to be the best tools in sculptured surface machining. In this study, a generic and unified kinematic model was developed as a viable alternative to the particular solutions that are only applicable to individual machine configurations. This versatile model is then used to verify the feasibility of the two rotational joints within the kinematic chain of three main types of a five-axis machine-tool. This versatile model is very useful applied to the design of five-axis machine tools.

Data inversion of multi-dimensional magnetic resonance in porous media

Since its inception in the 1970s,multi-dimensional magnetic resonance(MR)has emerged as a powerful tool for non-invasive investigations of structures and molecular interactions.MR spectroscopy beyond one dimension allows the study of the correlation,exchange processes,and separation of overlapping spectral information.The multi-dimensional concept has been re-implemented over the last two decades to explore molecular motion and spin dynamics in porous media.Apart from Fourier transform,methods have been developed for processing the multi-dimensional time-domain data,identifying the fluid components,and estimating pore surface permeability via joint relaxation and diffusion spectra.Through the resolution of spectroscopic signals with spatial encoding gradients,multi-dimensional MR imaging has been widely used to investigate the microscopic environment of living tissues and distinguish diseases.Signals in each voxel are usually expressed as multi-exponential decay,representing microstructures or environments along multiple pore scales.The separation of contributions from different environments is a common ill-posed problem,which can be resolved numerically.Moreover,the inversion methods and experimental parameters determine the resolution of multi-dimensional spectra.This paper reviews the algorithms that have been proposed to process multidimensional MR datasets in different scenarios.Detailed information at the microscopic level,such as tissue components,fluid types and food structures in multi-disciplinary sciences,could be revealed through multi-dimensional MR.

角色演变视角下辟谣信息对社交媒体意见领袖形成的影响——基于舆论领导法则

[目的/意义]探究辟谣信息对社交媒体用户的影响,特别是对意见领袖形成现象机制的挖掘。构建社交媒体用户角色演变行为模型,研究人员可以更好地了解意见领袖的形成原因,刻画辟谣信息对社交媒体用户的作用机制,有助于对参与辟谣的“准意见领袖”进行合理引导,以提高社交媒体舆情管理的效率和准确性。[方法/过程]首先,对微博社交媒体平台中的特定主题事件进行跟踪,识别并筛选出“新生意见领袖”群体,并利用质性访谈对此群体进行深度访谈。然后,基于程序化扎根理论方法,利用NVivo 12软件进行扎根编码构建模型,并通过理论饱和度检验验证模型的完整性。最后,对社交媒体用户角色演变的原因进行深入分析和探讨。[结果/结论]在辟谣信息作用下,个体主观因素、社会环境因素和辟谣信息因素对一般用户向新生意见领袖的演变行为产生重要影响,其中个体主观因素在辟谣信息的作用下对意见领袖形成的影响更加显著,自我认知清晰、高情绪需求且情绪稳定的个体更容易成为影响力持久的意见领袖;社会公众认可、实质性奖励和政策法律支持共同构成社会环境因素,是意见领袖形成的重要保障;辟谣信息因素中的辟谣信息热度、可信度和发布质量等方面都会直接影响“潜在意见领袖”用户群体对辟谣事件的把握,进而干预其参与辟谣的决定。

数字化转型与企业ESG表现——基于媒体关注度与高管过度自信的双重视角

基于2017—2022年沪深A股制造业上市公司的面板数据,从媒体关注度和高管过度自信两个视角出发,实证研究上市公司数字化转型影响ESG表现的两条路径。相关数据研究表明:第一,数字化转型与企业ESG表现存在显著的线性相关关系;第二,高管过度自信会影响数字化转型对企业ESG表现的促进作用;第三,数字化转型通过媒体关注度影响企业ESG表现。稳健性检验结果表明,在考虑倾向得分匹配法、工具变量法、替换被解释变量等方法后,数字化转型对企业ESG表现的促进作用仍然成立。异质性分析结果表明,国有企业和大规模企业能更好地利用数字化转型,积极履行社会责任,帮助企业提升ESG表现,为企业实现高质量发展提供了指引。

Transformer神经网络和轴心轨迹在燃机转子故障诊断中的应用

针对目前国内燃机电厂转子故障诊断主要依靠振动分析,无自动识别轴心轨迹设备的现状,将在自然语言处理领域大放异彩的Transformer神经网络引入转子轴心轨迹监测与识别领域,通过独立成分分析(Independent Component Analysis,ICA)与Transformer结合提升燃机转子故障诊断准确性。使用ICA代替Transformer自身的线性变换提取图片特征信息构建输入样本,为了解决自注意力机制无法捕捉位置信息的问题,提出相对位置编码方法,区别于使用较多的绝对位置编码,通过嵌入相对位置编码子层,将相对位置信息注入自注意力机制,使得Attention模块能够学习到序列的相对位置信息,可进一步提高图像识别模型的准确性,该模型对于轴心轨迹故障类型平均识别率达到93.8%。实验结果表明ICA-Transformer模型对转子轴心轨迹的识别准确率较高,对电厂运维具有一定的指导意义。

民营小微外贸企业数字化转型升级的内在逻辑和实现途径——以社交媒体赋能促进新质生产力形成视角

企业规模门槛效应阻碍了小微企业数字化转型升级中新质生产力的形成,尤其是民营小微外贸企业受规模限制普遍对数字化转型升级抱有“用不起、不会用、不愿用”的消极态度。近年来,国外消费者通过社交媒体进行购物的新消费模式正在形成。社交媒体具有低成本、高覆盖和精准定位的特点,易于改造成为跨境信息传播、营销推广的重要平台,为民营小微外贸企业数字化转型升级带来了新视野,提供了新机遇。本文针对民营小微外贸企业“两大、四缺乏”的现状,揭示了数字化转型升级带来的“四升一降”效应,并提出利用信息获取与反馈、市场拓展与推广、跨境交流与合作等机制,企业通过建立专业化团队、制定清晰策略、优化线上渠道、加强客户沟通等路径,创新出“社交媒体数字营销+小微外贸企业生产”深度融合的外贸新模式,培育新质生产力,促进传统产业转型升级和高质量发展。