LOCAL ORTHOGONAL TRANSFORMATION FOR ACOUSTIC WAVEGUIDE

In this paper, a local orthogonal transformation is created to transform the Helmholtz waveguide with curved interface to the one with a flat interface within the two layer medium, and the Helmholtz equation u xx +u zz +κ 2(x,z)u=0 is transformed to V +αV +β V +γV=0 . Numerical results demonstrate that the transformation is more feasible. This transformation is particularly useful for the research on wave propagation in acoustic waveguide.

结合CSWin-Transformer和门卷积的壁画图像修复方法

敦煌壁画是珍贵的文化遗产,但现存壁画存在着大量破损现象。针对现有图像修复方法在处理敦煌壁画时面临着计算复杂度高、纹理模糊和特征提取不足等问题,提出了一种结合CSWin-Transformer(cross stripe window-Transformer)和门卷积的壁画图像修复方法。构建由全局层网络和局部层门卷积残差密集网络组成的并行网络,利用条纹窗口增强图像特征提取能力,并通过门卷积残差块提升结构纹理修复的准确性。设计全局-局部特征融合模块来融合全局层和局部层输出的特征图像,以保持修复结果整体的一致性。通过建立共享注意力机制实现全局层和局部层之间的信息交互,同时为了完成破损壁画的修复,采用谱归一化马尔科夫判别模型进行对抗训练。通过对真实破损壁画的修复实验,结果表明,所提方法在主客观指标上均优于所对比的方法。

基于多层级视频Transformer的视觉自动定位方法

工业自动化产线中,设备的异常检测直接决定加工质量,由机械臂和搭载于机械臂前端的工业相机构成的视觉系统可以有效监测此类异常。本文使用六轴机械臂搭载工业相机对工件表面进行成像,获取由模糊到清晰再到模糊的视频序列,以此选出最清晰的视频帧作为自动加工中有聚焦要求的距离指导,以进行聚焦异常修正,从而实现自动定位。提出一种基于多层级视频Transformer的视频分类模型多级视频Transformer(MLVT)用于高语义级别的视频表征学习,并用于选出视频序列中成像最清晰的帧。首先,提出一种具有多种感受野的token划分方法多级标记(MLT),能够将原始视频数据按2D图像补丁、3D图像补丁、帧和片段这4个层级划分成token序列,并在加入位置编码之后送入多级编码器(MLE)方法进行注意力的计算。为了缓解多层级的tokens带来的计算代价和收敛速度慢的问题,MLE引入一种逐层的可变形注意力机制逐层可变形注意力机制(LWLA),以一种可学习的方式代替全局注意力进行特征相似性的计算。最终,该方法3个版本的模型在本文的视频数据集上分别取得了87.2%、88.6%、88.9%的分类准确率,在与同参数量级的主流视频Transformer实验对比中均表现了最优的性能,有效地完成了从视频序列中选择出最清晰帧的任务,能够为下游视觉任务的性能提供强有力保障。

多尺度局部特征和Transformer全局学习融合的发动机剩余寿命预测

飞机发动机剩余寿命(Remaining useful life,RUL)的准确预测对确保其安全性和可靠性至关重要.在基于多传感器检测数据预测时,需解决局部特征提取问题以全面捕捉设备在不同时间尺度下的退化趋势,并需解决时间序列中各元素之间长期依赖性的全局学习问题.因此,提出了结合多尺度局部特征增强单元(Multi-sacle local feature enhancement unit,MSLFU_BLOCK)和Transformer编码器的预测模型,称之为MS_Transformer.MSLFU_BLOCK利用堆叠的因果卷积逐层从时间序列数据中提取多尺度局部信息,同时避免了传统卷积计算中固有的未来数据泄漏问题.随后,Transformer编码器通过其自注意机制进一步捕获时间序列数据中的短期和长期依赖关系.通过将多尺度局部特征增强单元与Transformer编码器相结合,提出的MS_Transformer全面捕捉了时间序列数据中的局部和全局模式.在广泛使用的CMAPSS基准数据集上进行的消融和预测实验验证了模型的合理性和有效性.与13个先进预测模型的比较分析表明,MS_Transformer模型在操作条件更复杂的FD002和FD004数据集上的RMSE和Score指标优于其他模型,同时在四个数据集上的平均性能最优.该研究为发动机剩余寿命预测提供了更为可靠的解决方案.

LOCAL GOVERNMENT,ENTERPRISES AND INDIVIDUALS:ECONOMIC TRANSFORMATION IN THE PEARL RIVER DELTA—A Case Study in Beijiao Township,Shunde City

The Pearl River(Zhujiang)Delta(PRD)has been a focal point in reform era ac ademic circles not only for its dramatic industrial growth but a lso the simultaneous agricultural development.Unlike most of existing research on the PRDeconomic development and transformation fromthe whole region level,this paper explored this question f romthe perspec-tive of a township using Beijiao in Shunde City as a case study.Unlike the c onclusions of existing studies whic h attribute the regional economic transition to the macro factors,particularly the influence of external investment,t his research re-veals that at the level of township,t he local government,the town-villa ge owned enterprises and the individ uals have been playing remarkable roles in local economic transformation.In the early stage since the economic reform,Beijiao township government,replacing the central a nd provincial governments before,b egan to manipulate the development o f town-village owned enterprises and lead the local economic transformation from agricultural to industrial dominated.As the town-village owned enterprises grew during the later years,they gra dually acted as the main dominant pla yer leading the local agricultural and industrial growth.At the same time the individuals in Beijiao were playing more in dependent role to gain their most profits.While the local government changed to be t he real manager of local economies.S o the local economic transition was not entirely externally driven.In another word,the“driven from outsidemodel can not totally explain the economic fact in this specific region.

基于局部Transformer的泰语分词和词性标注联合模型

泰语分词和词性标注任务二者之间存在高关联性,已有研究表明将分词和词性标注任务进行联合学习可以有效提升模型性能,为此,提出了一种针对泰语拼写和构词特点的分词和词性标注联合模型。针对泰语中字符构成音节,音节组成词语的特点,采用局部Transformer网络从音节序列中学习分词特征;考虑到词根和词缀等音节与词性的关联,将用于分词的音节特征融入词语序列特征,缓解未知词的词性标注特征缺失问题。在此基础上,模型采用线性分类层预测分词标签,采用线性条件随机场建模词性序列的依赖关系。在泰语数据集LST20上的试验结果表明,模型分词F1、词性标注微平均F1和宏平均F1分别达到96.33%、97.06%和85.98%,相较基线模型分别提升了0.33%、0.44%和0.12%。

Comparison of Nonlinear Local Lyapunov Vectors with Bred Vectors, Random Perturbations and Ensemble Transform Kalman Filter Strategies in a Barotropic Model

The breeding method has been widely used to generate ensemble perturbations in ensemble forecasting due to its simple concept and low computational cost. This method produces the fastest growing perturbation modes to catch the growing components in analysis errors. However, the bred vectors （BVs） are evolved on the same dynamical flow, which may increase the dependence of perturbations. In contrast, the nonlinear local Lyapunov vector （NLLV） scheme generates flow-dependent perturbations as in the breeding method, but regularly conducts the Gram-Schmidt reorthonormalization processes on the perturbations. The resulting NLLVs span the fast-growing perturbation subspace efficiently, and thus may grasp more com- ponents in analysis errors than the BVs. In this paper, the NLLVs are employed to generate initial ensemble perturbations in a barotropic quasi-geostrophic model. The performances of the ensemble forecasts of the NLLV method are systematically compared to those of the random pertur- bation （RP） technique, and the BV method, as well as its improved version--the ensemble transform Kalman filter （ETKF） method. The results demonstrate that the RP technique has the worst performance in ensemble forecasts, which indicates the importance of a flow-dependent initialization scheme. The ensemble perturbation subspaces of the NLLV and ETKF methods are preliminarily shown to catch similar components of analysis errors, which exceed that of the BVs. However, the NLLV scheme demonstrates slightly higher ensemble forecast skill than the ETKF scheme. In addition, the NLLV scheme involves a significantly simpler algorithm and less computation time than the ETKF method, and both demonstrate better ensemble forecast skill than the BV scheme.

Matrix Tensor Product Approach to the Equivalence of Multipartite States under Local Unitary Transformations

The equivalence of multipartite quantum mixed states under local unitary transformations is studied. A criterion for the equivalence of non-degenerate mixed multipartite quantum states under local unitary transformations is presented.

Crystalline Morphology and Local Transformation of C_(60) Molecular Crystal

1 Introduction C60 crystal is a new form of pure, solid carbon and has been widely investigated. But until now, there still exists the ambiguity in its crystal structure. Early paper showed that C60 molecules stack in a hexagonal close-packed （hcp） lattice with α=10.02A, c=16.36A, but other ones reported that C60 crystal has a face-centred-cubic （foe） lattice with α=14.172,. Foc and hcp structures are both stable at room temperature, whereas from the energetic point of view, fcc structure is much stabler. Moreover, the differences in the

Adaptive Biorthogonal Local Discrete Cosine Transform for Interference Excision in Direct Sequence Spread Spectrum Communications

A novel time-frequency domain interference excision technique is proposed. The technique is based on adaptive biorthogonal local discrete cosine trans form (BLDCT). It uses a redundant library of biorthogonal local discrete cosine bases and an efficient concave cost function to match the transform basis to the interfering signal. The main advantage of the algorithm over conventional trans form domain excision algorithms is that the basis functions are not fixed but ca n be adapted to the time-frequency structure of the interfering signal. It is w e ll suited to transform domain compression and suppression of various types of in terference. Compared to the discrete wavelet transform (DWT) that provides logar ithmic division of the frequency bands, the adaptive BLDCT can provide more flex ible frequency resolution. Thus it is more insensitive to variations of jamming frequency. Simulation results demonstrate the improved bit error rate (BER) perf ormance and the increased robustness of the receiver.

基于Local-Global-VIT细粒度分类算法的蝴蝶识别

【目的】准确鉴别蝴蝶种类,动态观测蝴蝶群落多样性变化对生境质量评估、生态环境恢复等方面具有重要意义。针对现有蝴蝶识别方法仅依靠整体特征,忽略了局部特征导致识别生态图像能力不足的问题,本研究旨在开发一种Local-Global-VIT细粒度分类算法的蝴蝶识别方法。【方法】本研究以5科200种共计25 279张蝴蝶图像为识别对象,采用多种数据增强方法扩充图像数据;通过视觉Transformer(vision transformer, VIT)层级结构及自注意力机制逐层选择局部令牌并保留至最后一层学习蝴蝶局部判别部位信息;聚合高层全局令牌消除复杂背景干扰;通过对比损失拉大类间距提高区分度。除此之外,使用合理的学习率调整策略和迁移学习方法,优化了模型收敛过程,在不增加参数量的情况下提高了性能。【结果】Local-Global-VIT算法在大规模细粒度公开数据集Butterfly-200上识别准确率达91.20%,较改进前提升了1.15%,比最优的一般害虫识别算法EfficientNet_b0和细粒度分类算法TransFG准确率分别高了1.83%和0.64%,F1分值分别提高了1.89%和0.88%。【结论】Local-Global-VIT算法以细粒度识别方式有效解决了蝴蝶类内差异大、类间差异小的分类难题,能准确地识别蝴蝶种类,有助于高效评估生境质量。

融合Transformer和交互注意力网络的方面级情感分类模型

现有的大多数研究者使用循环神经网络与注意力机制相结合的方法进行方面级情感分类任务。然而,循环神经网络不能并行计算,并且模型在训练过程中会出现截断的反向传播、梯度消失和梯度爆炸等问题,传统的注意力机制可能会给句子中重要情感词分配较低的注意力权重。针对上述问题,该文提出了一种融合Transformer和交互注意力网络的方面级情感分类模型。首先利用BERT(bidirectional encoder representation from Transformers)预训练模型来构造词嵌入向量,然后使用Transformer编码器对输入的句子进行并行编码,接着使用上下文动态掩码和上下文动态权重机制来关注与特定方面词有重要语义关系的局部上下文信息。最后在5个英文数据集和4个中文评论数据集上的实验结果表明,该文所提模型在准确率和F1上均表现最优。

改进Transformer的高光谱图像地物分类方法——以黄河三角洲为例

高光谱技术已成为沿海湿地监测的主要手段,但传统高光谱分类方法通常存在特征提取不充分、同物异谱和场景碎片化等问题。针对这些问题,该文将Transformer用于高光谱分类,提出一种新的分类方法。该方法基于视觉自注意力模型(Vision Transformer,ViT),利用Non-local技术学习全局空间特征,扩大感受野解决提取判别特征不足的问题;同时,通过自适应跨层残差连接加强层间信息交换,解决信息损失的问题。选取NC16和NC13黄河三角洲湿地数据集作为实验数据,并将提出的方法与支持向量机(support vector machine,SVM)、一维卷积神经网络(one dimensional convolution neural network,1DCNN)、上下文深度卷积神经网络(contextual deep convolution neural network,CDCNN)、光谱空间残差网络(spectral-spatial residual network,SSRN)、混合光谱网络(hybrid spectral network,HybridSN)和ViT进行比较分析。结果表明,所提方法的总体精度(overall accuracy,OA)、平均精度(average accuracy,AA)和Kappa系数均有显著提高,OA分别达到96.24%和73.84%,AA分别达到83.42%和74.87%,Kappa分别达到94.80%和68.94%。

基于改进Vision Transformer网络的农作物病害识别方法

基于DCNN模型的农作物病害识别方法在实验室环境下识别准确率高,但面对噪声时缺少鲁棒性.为了兼顾农作物病害识别的精度和鲁棒性,本文在标准ViT模型基础上加入增强分块序列化和掩码多头注意力,解决标准ViT模型缺乏局部归纳偏置和视觉特征序列的自注意力过于关注自身的问题.实验结果表明,本文的EPEMMSA-ViT模型对比标准ViT模型可以更高效的从零学习;当添加预训练权重训练网络时,EPEMMSA-ViT模型在数据增强的PlantVillage番茄子集上能够得到99.63%的分类准确率;在添加椒盐噪声的测试数据集上,对比ResNet50、DenseNet121、MobileNet和ConvNeXt的分类准确率分别提升了6.08%、9.78%、29.78%和12.41%;在添加均值模糊的测试数据集上,对比ResNet50、DenseNet121、MobileNet和ConvNeXt的分类准确率分别提升了18.92%、31.11%、20.37%和19.58%.

Strategies for Localized Transformation of Landscape in Old Industrial Areas: A Case Study of Xiadian Industrial Area in Xuzhou City

Based on the theory of “localization”, the landscape status of Xiadian industrial area in Xuzhou City was investigated and analyzed. Localized transformation of landscape in the old industrial area can be conducted from the restoration of landscape ecological environment, protection of industrial landscape heritage, and sustainable utilization of industrial waste resources. It can achieve a better balance between urban renewal and the landscape transformation of the old industrial area and then realize the reshaping and regeneration of landscape and promote the development of local industries and the continuation of industrial culture to provide useful thinking for creating geographically representative urban landscape.

基于SF-Transformer的智能教育平台短期电力负荷预测研究

建设智能教育平台是推动教育智能化的一个重要过程,但智能教育平台依赖的人工智能模型在训练过程中会消耗大量电力,因此,开展短期电力负荷预测对建设智能教育平台具有重要意义.针对在考虑多个属性开展短期电力负荷预测时,由于部分属性与电力负荷数据的相关性不强并且Transformer无法捕捉电力负荷数据的时间相关性,而导致电力负荷预测不够准确的问题,基于SR(Székely and Rizzo)距离相关系数、融合时间定位编码和Transformer,提出了一种短期电力负荷预测模型SF-Transformer.SF-Transformer通过SR距离相关系数对影响电力负荷数据的属性进行筛选,选择与电力负荷数据之间SR距离相关系数较大的属性.SF-Transformer采用一种全局时间编码与局部位置编码相结合的融合时间定位编码,有助于模型全面获取电力负荷数据的时间定位信息.在数据集上开展了实验,实验结果表明SF-Transformer与其他模型相比,在两种时长上进行电力负荷预测具有更低的均方根误差和平均绝对误差.

Accurate Registration of Remote Sensing Images Based on Local Optimal Transformation

As the basic work of image stitching and object recognition,image registration played an important part in the image processing field.Much previous work in registration accuracy and realtime performance progressed very slowly,especially in registrating images with line feature.An innovative method for image registration based on lines is proposed,it can effectively improve the accuracy and real-time performance of image registration.The line feature can deal with some registration problems where point feature does not work.Our registration process is divided into two parts.The first part determines the rough registration transformation relation between reference image and test image.Then the similarity degree among different transformation and modified nonmaximum suppression(MNMS)algorithms are obtained,which produce local optimal solution to optimize the rough registration transformation.The final optimal registration relation can be obtained from two registration parts according to the match scores.The experimental results show that the proposed method makes a more accurate registration relation and performs better in real-time situation.

LOCALIZED RADON-WIGNER TRANSFORM AND GENERALIZED-MARGINAL TIME-FREQUENCY DISTRIBUTIONS

This paper introduces the localized Radon transform (LRT) into time-frequency distributions and presents the localized Radon-Wigner transform (LRWT). The definition of LRWT and a fast algorithm is derived, the properties of LRWT and its relationship with Radon-Wigner transform, Wigner distribution (WD), ambiguity function (AF), and generalized-marginal time-frequency distributions are analyzed.

宽卷积局部特征扩展的Transformer网络故障诊断模型

视觉Transformer网络的高精度诊断性能依赖于充分的训练数据,利用卷积网络在提取局部特征上的优势,构造能同时描述故障局部和全局特征的提取层,提高诊断模型的抗噪声干扰能力。首先,引入卷积网络模块将原始振动信号转换为Transformer网络可以直接接收的特征向量,提取故障局部特征,并通过增加卷积网络的感受野。然后,结合Transformer网络多头自注意力机制生成的全局信息,构建能同时描述故障局部和全局特征的特征向量。最后,在Transformer网络的预测层,利用高效通道注意力机制对特征向量的贡献度进行自动筛选。在西储大学(CWRU)轴承数据集上的故障诊断结果表明,在信噪比-4 dB的噪声干扰下,改进后的Transformer网络轴承故障诊断模型的准确率达90.21%,与原始Transformer模型相比,准确率提高了13.2%,在噪声环境下表现出优异的诊断性能。

USING WAVELET TRANSFORM TO STUDY THELIPSCHITZ LOCAL SINGULAR EXPONENTIN WALL TURBULENCE

In this paper, wavelet,transform is introduced to study the Lipschitz local singular exponent for characterising the local singularity behavior of fluctuating velocity in wall turbulence. I, is found that the local singular exponent is negative when the ejections and sweeps of coherent structures occur in a turbulent boundary layer.