基于多模态融合和自适应剪枝Transformer的脑肿瘤图像分割算法

脑肿瘤是目前世界上最致命的肿瘤之一,所以脑肿瘤图像的自动分割在临床诊疗中变得日益重要.近年来,基于CNN和Transformer的脑肿瘤分割方法在医学图像分割领域取得了令人欣喜的成就.然而,大多数方法没有充分利用脑肿瘤多模态间的互补性和差异性,并且模型中的Transformer在捕获远程依赖性的同时,忽略了其较大的计算复杂性、冗余依赖性等问题.针对此问题,提出一种基于多模态融合和自适应剪枝Transformer的脑肿瘤图像分割方法(MF-MAPT Swin UNETR),其中多模态融合模块可以充分学习性质相近的模态间信息和不同模态不同尺度的特征变化,为后续分割提供了充分的准备;基于多模态的自适应剪枝Transformer可以降低计算复杂度,对提升性能有一定的帮助,将MF-MAPT Swin UNETR模型在两个公共数据集上进行了实验验证,结果表明,该模型较最先进的方法整体具有突出的分割性能.

Over Voltage Fault due to Disconnection of Consumer’s Transformer Neutral Wire

Practically,the load currents in three phases are asymmetric in the power system.It means that the impedances are different in all three phases.If the consumer’s transformer neutral cut off and/or was disconnected from the neutral of power supply source,then there will be some trouble and failure occurred.The current in the neutral wire drops down to zero when the neutral wire is cut off and the phase currents of all three-phase equal to each other since there was no return wire.The currents are equal but the voltages at the phase consumers are different.Especially for residential single-phase consumers,the voltage at the consumers of the phase varies differently for three phase systems when the neutral wire was disconnected at consumer side and even the voltage at the consumers one or two of those three phases becomes over nominal voltage or reaches nearly line voltage.In this case,the electronic appliances in that phase will be fed by high voltage than the rated value and they can be broken down.In the power system of UB(Ulaanbaatar)city,there are some occasional such kind of failures every year.Obviously,many electronic appliances were broken down due to high voltage and the electricity utility companies respond for service charge of damaged parts.

Positron Emission Tomography Lung Image Respiratory Motion Correcting with Equivariant Transformer

In addressing the challenge of motion artifacts in Positron Emission Tomography (PET) lung scans, our studyintroduces the Triple Equivariant Motion Transformer (TEMT), an innovative, unsupervised, deep-learningbasedframework for efficient respiratory motion correction in PET imaging. Unlike traditional techniques,which segment PET data into bins throughout a respiratory cycle and often face issues such as inefficiency andoveremphasis on certain artifacts, TEMT employs Convolutional Neural Networks (CNNs) for effective featureextraction and motion decomposition.TEMT’s unique approach involves transforming motion sequences into Liegroup domains to highlight fundamental motion patterns, coupled with employing competitive weighting forprecise target deformation field generation. Our empirical evaluations confirm TEMT’s superior performancein handling diverse PET lung datasets compared to existing image registration networks. Experimental resultsdemonstrate that TEMT achieved Dice indices of 91.40%, 85.41%, 79.78%, and 72.16% on simulated geometricphantom data, lung voxel phantom data, cardiopulmonary voxel phantom data, and clinical data, respectively. Tofacilitate further research and practical application, the TEMT framework, along with its implementation detailsand part of the simulation data, is made publicly accessible at https://github.com/yehaowei/temt.

联合自适应核和Transformer的脊柱磁共振成像多类别分割网络

针对脊柱磁共振成像(magnetic resonance imaging,MRI)结构复杂,存在多余组织、噪声及伪像的问题,本研究设计了一种联合自适应核和Transformer的脊柱MRI实例多类别分割网络。以Swin Transformer作为骨干网络,通过引入稠密连接模块减少前向通道的信息丢失,以更好地捕获图像中的细节和局部信息。同时,为进一步捕获复杂空间的多尺度特征,采取自注意力核选择的方式构建跨尺度稠密连接,使模型在训练过程中能自适应学习到合适的卷积核尺寸,提高模型对不同尺度信息的感知能力,提高分割性能。通过在215例受试者的T2加权MRI图像2D切片上进行验证,实验结果显示,该算法的平均交并比(mean intersection over union,mIoU)、平均召回率(mean recall rate,mRecall)和平均骰子系数(mean dice coefficient,mDice)分别为82.63%、89.37%和88.85%。结果表明,本研究算法的分割性能较好,可实现脊柱MRI中椎体及椎间盘的精准分割,为临床医生提供辅助诊断工具。

改进的密集视频描述Transformer译码算法

当Transformer应用于密集视频描述时,历史文本特征会对后续文本生成产生干扰,难以捕捉视频动态信息从而影响描述的连贯性和准确性。为保持上下文一致性的同时又能缓解历史文本干扰,提出改进的密集视频描述Transformer译码算法(D-Uformer)。该算法利用前馈神经网络(FNN)增强历史文本特征表达,通过跳跃连接构建删除冗余支路和增强补足支路,利用减法降低历史文本特征过度聚焦导致描述不准确的影响,提高模型对输入视频特征的关注度;同时,利用加法弥补特征传递过程中丢失的上下文信息,生成准确且连贯表达当前视频内容的描述语句。在ActivityNet和Charades数据集上的实验结果表明,D-Uformer算法的描述性能提升明显,与视频多样性描述网络(TDPC)相比,准确性最高提升4.816%,多样性最高提升4.167%,生成的描述不仅更贴合视频内容,且更符合人类语言习惯。

Study Hybrid Compensation Cophase Traction Power Supply System Schemes for High-Speed and Heavy Haul Electrical Railway with V Connection Transformer

In order to solve negative phase sequence problem of V connection transformer in the high speed and heavy haul electrical railway of China, the hybrid compensative co-phase traction power supply system which based on passive and active compensation is proposed. Firstly, There construction and capacity distribution are analyzed, and the compensation current of active equipment is gave;Second, the feature of the hybrid compensative schemes are discussed. In the end, the related simulation results have confirmed the effectiveness of the compensation schemes in this paper.

Brain Functional Network Generation Using Distribution-Regularized Adversarial Graph Autoencoder with Transformer for Dementia Diagnosis

The topological connectivity information derived from the brain functional network can bring new insights for diagnosing and analyzing dementia disorders.The brain functional network is suitable to bridge the correlation between abnormal connectivities and dementia disorders.However,it is challenging to access considerable amounts of brain functional network data,which hinders the widespread application of data-driven models in dementia diagnosis.In this study,a novel distribution-regularized adversarial graph auto-Encoder(DAGAE)with transformer is proposed to generate new fake brain functional networks to augment the brain functional network dataset,improving the dementia diagnosis accuracy of data-driven models.Specifically,the label distribution is estimated to regularize the latent space learned by the graph encoder,which canmake the learning process stable and the learned representation robust.Also,the transformer generator is devised to map the node representations into node-to-node connections by exploring the long-term dependence of highly-correlated distant brain regions.The typical topological properties and discriminative features can be preserved entirely.Furthermore,the generated brain functional networks improve the prediction performance using different classifiers,which can be applied to analyze other cognitive diseases.Attempts on the Alzheimer’s Disease Neuroimaging Initiative(ADNI)dataset demonstrate that the proposed model can generate good brain functional networks.The classification results show adding generated data can achieve the best accuracy value of 85.33%,sensitivity value of 84.00%,specificity value of 86.67%.The proposed model also achieves superior performance compared with other related augmentedmodels.Overall,the proposedmodel effectively improves cognitive disease diagnosis by generating diverse brain functional networks.

基于Transformer的核磁共振肿瘤图像分割研究

为了提高医务人员对于脑肿瘤核磁共振成像的诊断效率和准确率,提出一种基于Transformer的U形图像分割网络,使用纯Transformer架构对输入图像进行特征编码与解码,并通过跳转连接融合多尺度特征修复目标细节。此外,还在特征融合的地方增加了跳转注意力操作,增强模型的鲁棒性。为了验证此方法的有效性,在公共的脑肿瘤分割数据集BraTS2021上进行了多次实验。所提模型对ET、TC和WT区域的分割Dice得分分别达到83.51%、87.66%和91.39%,且参数量和浮点运算次数较低。实验结果表明,所提网络结构在脑肿瘤核磁共振成像上具有较强的分割性能。

基于CNN‐Head Transformer编码器的中文命名实体识别

基于多头自注意力机制的Transformer作为主流特征提取器在多种自然语言处理任务中取得了优异表现,但应用于命名实体识别任务时存在一字多词、增加额外存储与词典匹配时间等问题。提出一种CNN-Head Transformer编码器(CHTE)模型,在未使用外部词典和分词工具的基础上,通过自注意力机制捕获全局语义信息,利用不同窗口大小的CNN获取Transformer中6个注意力头的Value向量,使CHTE模型在保留全局语义信息的同时增强局部特征和潜在词信息表示,并且应用自适应的门控残差连接融合当前层和子层特征,提升了Transformer在命名实体识别领域的性能表现。在Weibo和Resume数据集上的实验结果表明,CHTE模型的F1值相比于融合词典信息的Lattice LSTM和FLAT模型分别提升了3.77、2.24和1.30、0.31个百分点,具有更高的中文命名实体识别准确性。

U-shaped Vision Transformer and Its Application in Gear Pitting Measurement

Although convolutional neural networks have become the mainstream segmentation model,the locality of convolution makes them cannot well learn global and long-range semantic information.To further improve the performance of segmentation models,we propose U-shaped vision Transformer(UsViT),a model based on Transformer and convolution.Specifically,residual Transformer blocks are designed in the encoder of UsViT,which take advantages of residual network and Transformer backbone at the same time.What is more,transpositions in each Transformer layer achieve the information interaction between spatial locations and feature channels,enhancing the capability of feature learning.In the decoder,for enhancing receptive field,different dilation rates are introduced to each convolutional layer.In addition,residual connections are applied to make the information propagation smoother when training the model.We first verify the superiority of UsViT on automatic portrait matting public dataset,which achieves 90.43%accuracy(Acc),95.56%Dice similarity coefficient,and 94.66%Intersection over Union with relatively fewer parameters.Finally,UsViT is applied to gear pitting measurement in gear contact fatigue test,and the comparative results indicate that UsViT can improve the Acc of pitting detection.

Effect of pile-cap connection on behavior of torsionally loaded pile groups

To evaluate the responses of fixed and pinned pile groups under torsion, a method is presented to analyze the nonlinear behavior of free-standing pile groups with rigid pile caps. The method is capable of simulating the nonlinear soil response in the near field usingp-y and r-θ curves, the far-field interactions through Mindlin＇s and Randolph＇s elastic solutions, and the coupling effect of lateral resistance on torsional resistance of the individual piles using an empirical factor. Based on comparisons of the solutions for fixed- and pinned-head, 1×2, 2×2, and 3×3 pile groups subjected to torsion, it was found that pile-cap connection significantly influences the torsional capacity of pile groups and the assignment of applied torques in the pile groups. In this study, the applied torques for the pinned-head pile groups are only 44%-64% of those for the corresponding fixed-head pile groups at a twist angle of 2^o. Such a difference is mainly due to the change of the lateral resistances of individual piles in the groups.

Wind turbine clutter mitigation using morphological component analysis with group sparsity

To address the problem that dynamic wind turbine clutter(WTC)significantly degrades the performance of weather radar,a WTC mitigation algorithm using morphological component analysis(MCA)with group sparsity is studied in this paper.The ground clutter is suppressed firstly to reduce the morphological compositions of radar echo.After that,the MCA algorithm is applied and the window used in the short-time Fourier transform(STFT)is optimized to lessen the spectrum leakage of WTC.Finally,the group sparsity structure of WTC in the STFT domain can be utilized to decrease the degrees of freedom in the solution,thus contributing to better estimation performance of weather signals.The effectiveness and feasibility of the proposed method are demonstrated by numerical simulations.

亲清政商关系与企业数字化转型

在数字经济时代,数字化转型是企业实现高质量发展的内生要求,而企业的数字化转型离不开政府的引导和支持。基于此,分析检验当前亲清政商关系构建会如何影响微观企业的数字化转型,有助于企业实现高质量发展。利用2016—2020年中国A股传统行业上市公司的年度观察数据研究发现,亲清政商关系对企业数字化转型具有显著的正向影响,即在政商关系越亲清的地区,企业的数字化水平越高,并且这种正向影响在非国有企业中表现得更强。进一步的机制检验表明,亲清政商关系主要通过“市场导向效应”和“资源供给效应”对企业数字化转型发挥积极的促进作用。更具有启发意义的是,主要是政商关系的“亲近化”而不是政商关系的“清白化”助力了企业的数字化转型,而且亲清政商关系能够降低传统政治关联的相关影响。

基于tr-PCBAMSiam的小目标跟踪算法

无人机在目标跟踪过程中,存在分辨率低、运动模糊、目标遮挡、目标密集、相似目标干扰等问题,导致算法跟踪精度下降。针对这一问题,在SiamRPN的基础上提出tr-PCBAMSiam,即一种基于混合注意力的聚合残差连接、transformer互相关运算以及基于无锚框的区域回归网络的目标跟踪算法。将该算法与其他目标跟踪算法在OTB100数据集上进行对比,在精度和成功率方面,与SiamRPN算法相比分别有6.9%和8%的提升;在LaTOT数据集上与SiamRPN相比,精度和成功率分别有13.1%和8.5%的提升。

基于PCSA-YOLOv7 Former的输电线路连接金具及其锈蚀检测方法

输电线路分布情况复杂且故障难以有效检测,其中连接金具长期暴露于复杂环境下易受到恶劣环境的影响出现锈蚀等故障。针对输电线路连接金具部件具有尺度多样性和存在着锈蚀故障检测精度低的问题,提出了一种基于双重注意力嵌入重构和Swin Transformer的输电线路连接金具组件及其锈蚀故障检测方法:PCSA-YOLOv7 Former。实验结果表明:该方法在构建的TLCF数据集上的综合检测性能领先于12类当前先进的目标检测算法,其中在测试集上的mAP_(0.5)达到94.9%,该方法相比于基线模型YOLOv7,其F1和mAP0.5指标分别提升了2.6个百分点和2.2个百分点,说明该方法能够更全面地理解输电线路连接金具图像中的多尺度语义信息并学习到不易区分的微小细节表征。

基于DH-Swin Unet的医学图像分割算法

骨关节疾病自古以来是人类最高发的疾病之一,随着老龄化的不断加快,这类疾病日趋广泛,关节外科医师面临着巨大挑战.对人体关节的图像分割方法研究可以帮助医生进行临床诊断和治疗,然而,由于存在噪声、模糊、对比度低等问题,医学图像的特征提取比普通图像更具挑战性,而且目前大多数分割模型在编码器和解码器之间都采用了普通的跳跃连接,没有注重解决跳跃连接过程中的信息间隙和损失问题.为解决这些问题,提出一种基于DH-Swin Unet的医学图像分割算法,该模型在Swin-Unet模型的基础上,在跳跃连接中引入密集连接的Swin Transformer块,并加入混合注意力机制,来强化网络的特征信息传递.通过在某三甲医院提供的真实临床数据对所提方法的性能进行评价,结果表明,所提出的方法取得了DSC为86.79%、HD为32.05 mm的分割结果,在关节疾病的临床诊断中具有一定的实用价值.

多通道电流信号深度特征融合的开关磁阻电机故障诊断研究

提出一种基于卷积神经网络(convolution neural networks,CNN)与Transformer结合的新型深度学习框架1D-Uniformer(one dimensional unified transformer),解决开关磁阻电机高阻接触故障和相间短路故障的分类识别问题。搭建开关磁阻电机故障诊断实验平台,在开关磁阻电机定子绕组上设置高阻接触故障和相间短路故障,并通过非侵入式方法采集电机的三相电流信号;引入动态位置嵌入、多头关系聚合器和前馈网络对传统CNN进行改进,得到1D-Uniformer以充分提取高阻接触故障和相间短路故障的特征。实验结果表明:该模型在高阻接触故障和相间短路故障诊断方面均具有很好的分类效果,在18种故障状态下识别精度能达到100%,在不同的噪声强度下仍然具有较高的鲁棒性。

基于数智化视角的企业集团智慧型司库建设探究

随着数字经济的快速发展,企业集团需要加快数智化转型,作为集团运营管理重要载体的司库也需要相应转型。为满足司库数智化转型的三个基本要求。首先,需从司库的操作运营层、资源管控层、决策分析层等三个层级着手创新设计司库数智化转型的功能架构。其次,为支撑该功能架构,需从先进性、集成性、安全性、可拓展性等四个方面着手创新设计司库数智化转型的技术机构。最后,为确保建成智慧型司库,需从机制健全、过程管控、标准优化、数据治理等四个关键环节发力提供坚强保障举措。

肌源干细胞通过TGF-β1/CTGF信号通路减轻压力性尿失禁大鼠的排尿障碍和尿道损伤

目的基于转化生长因子β1/结缔组织生长因子(TGF-β1/CTGF)信号通路探究肌源干细胞(MDSCs)对压力性尿失禁(SUI)大鼠排尿功能和尿道损伤的影响。方法分离、培养MDSCs,免疫细胞化学法检测MDSCs标志物Desmin和Sca-1表达。40只SD雌性大鼠按照随机数字表法分为4组:NC组(不作任何处理)、SUI组(SUI模型)、MDSCs组(SUI模型+MDSCs治疗)、MDSCs+SB431542组(SUI模型+MDSCs治疗+通路抑制剂SB431542处理),每组10只。比较4组大鼠喷嚏阳性率、尿动力学指标[膀胱最大容积(MBC)、漏尿点压(LPP)、腹压漏尿点压(ALPP)、排尿量、残余尿量、排尿效率];HE染色观察4组尿道组织病理学变化;蛋白免疫印迹法检测4组尿道组织中TGF-β1、CTGF蛋白表达。结果成功获得Desmin和Sca-1均呈阳性表达的MDSCs。与NC组比较,SUI组喷嚏阳性率、MBC、LPP、ALPP、排尿量、残余尿量均增加,排尿效率、TGF-β1和CTGF蛋白表达均降低(P<0.05),伴有严重尿道损伤;与SUI组比较,MDSCs组喷嚏阳性率、MBC、LPP、ALPP、排尿量、残余尿量均降低,排尿效率、TGF-β1和CTGF蛋白表达均增加(P<0.05),尿道损伤减轻;与MDSCs组比较,MDSCs+SB431542组喷嚏阳性率、MBC、LPP、ALPP、排尿量、残余尿量均增加,排尿效率、TGF-β1和CTGF蛋白表达均降低(P<0.05),尿道损伤加重。结论MDSCs可改善SUI大鼠的排尿功能,减轻尿道损伤,其可能是通过调控TGF-β1/CTGF信号通路发挥作用。

企业数字化转型促进了资本跨区域流动吗——基于异地并购视角的研究

加快资本跨区域流动对构建以国内大循环为主体、国内国际双循环相互促进的新发展格局具有重要的战略意义。本文以2008—2022年中国A股上市公司为样本,借助异地并购这一重要的资本跨区域流动形式,考察新发展格局下企业数字化转型对资本跨区域流动的影响。研究发现:企业数字化转型通过降低并购交易成本和提高集团管控能力,有效地促进了异地并购决策,尤其对跨省并购决策的促进作用更显著。异质性分析结果表明,当并购企业所在地区市场化程度更高、税负水平更低、数字金融发展水平更高以及目标企业所在地区营商环境更好时,企业数字化转型对异地并购决策的促进作用更强;同时,这一促进作用在民营企业、非高新技术企业以及高管具有金融背景或海外背景的企业中更显著。本研究丰富了数字经济时代资本跨区域流动驱动因素的研究,从异地并购视角提供了企业数字化转型促进资本跨区域流动的微观证据。