In this paper, equivalent circuits for high frequency multi-winding magnetic components are derived from finite element (FE) computations. Lumped parameter models are first presented, based on previously published w...In this paper, equivalent circuits for high frequency multi-winding magnetic components are derived from finite element (FE) computations. Lumped parameter models are first presented, based on previously published work. All parameters of these circuits can be interpreted as the results of open and short-circuit tests on the transformer. Based on this consideration, numerical procedures are then proposed to derive frequency-dependent lumped parameters from FE simulations. By using an adequate formulation, parameters are directly obtained from the FE model degrees of freedom, without performing any volume integration in post-processing, which can be source of numerical errors. In this contribution, attention is paid on the modeling of magnetic coupling using inductances, and dissipative effects (winding and core losses) using resistances. The impact of conductor eddy currents on the circuit parameters is moreover studied in details. Instead of an analysis of the impact conductor eddy currents may have on the circuit parameters is moreover carried through.展开更多
针对主流Transformer网络仅对输入像素块做自注意力计算而忽略了不同像素块间的信息交互,以及输入尺度单一导致局部特征细节模糊的问题,本文提出一种基于Transformer并用于处理视觉任务的主干网络ConvFormer. ConvFormer通过所设计的多...针对主流Transformer网络仅对输入像素块做自注意力计算而忽略了不同像素块间的信息交互,以及输入尺度单一导致局部特征细节模糊的问题,本文提出一种基于Transformer并用于处理视觉任务的主干网络ConvFormer. ConvFormer通过所设计的多尺度混洗自注意力模块(Channel-Shuffle and Multi-Scale attention,CSMS)和动态相对位置编码模块(Dynamic Relative Position Coding,DRPC)来聚合多尺度像素块间的语义信息,并在前馈网络中引入深度卷积提高网络的局部建模能力.在公开数据集ImageNet-1K,COCO 2017和ADE20K上分别进行图像分类、目标检测和语义分割实验,ConvFormer-Tiny与不同视觉任务中同量级最优网络RetNetY-4G,Swin-Tiny和ResNet50对比,精度分别提高0.3%,1.4%和0.5%.展开更多
遥感图像的道路分割任务是遥感应用领域的一个研究热点,一直受到广泛的关注。由于遥感图像天然具备背景复杂、目标密集等特性,全局语义信息的构建对于准确提取遥感图像中道路是至关重要的。因此,基于Transformer模型进行优化,提出了基...遥感图像的道路分割任务是遥感应用领域的一个研究热点,一直受到广泛的关注。由于遥感图像天然具备背景复杂、目标密集等特性,全局语义信息的构建对于准确提取遥感图像中道路是至关重要的。因此,基于Transformer模型进行优化,提出了基于空间可分离注意力的跨尺度令牌嵌入Transformer遥感道路提取模型Cross-RoadFormer。具体而言,针对图像中道路尺度不统一的问题,设计了跨尺度编码层,将不同尺度的特征编码作为一个令牌嵌入整体,作为Transformer的输入,解决了Transformer跨尺度交互的问题;此外,提出了一种空间可分离注意力,其中,局部分组注意力获取细粒度、短距离信息,全局采样注意力捕获长距离、全局上下文信息,在保证道路提取准确度的前提下,降低了模型的计算量。在Massachusetts数据集和DeepGlobe数据集上的实验表明,提出的Cross-RoadFormer都实现了更高的IoU(intersection over union),分别为68.40%和58.04%,展现了该方法的优越性。展开更多
文摘In this paper, equivalent circuits for high frequency multi-winding magnetic components are derived from finite element (FE) computations. Lumped parameter models are first presented, based on previously published work. All parameters of these circuits can be interpreted as the results of open and short-circuit tests on the transformer. Based on this consideration, numerical procedures are then proposed to derive frequency-dependent lumped parameters from FE simulations. By using an adequate formulation, parameters are directly obtained from the FE model degrees of freedom, without performing any volume integration in post-processing, which can be source of numerical errors. In this contribution, attention is paid on the modeling of magnetic coupling using inductances, and dissipative effects (winding and core losses) using resistances. The impact of conductor eddy currents on the circuit parameters is moreover studied in details. Instead of an analysis of the impact conductor eddy currents may have on the circuit parameters is moreover carried through.
文摘针对主流Transformer网络仅对输入像素块做自注意力计算而忽略了不同像素块间的信息交互,以及输入尺度单一导致局部特征细节模糊的问题,本文提出一种基于Transformer并用于处理视觉任务的主干网络ConvFormer. ConvFormer通过所设计的多尺度混洗自注意力模块(Channel-Shuffle and Multi-Scale attention,CSMS)和动态相对位置编码模块(Dynamic Relative Position Coding,DRPC)来聚合多尺度像素块间的语义信息,并在前馈网络中引入深度卷积提高网络的局部建模能力.在公开数据集ImageNet-1K,COCO 2017和ADE20K上分别进行图像分类、目标检测和语义分割实验,ConvFormer-Tiny与不同视觉任务中同量级最优网络RetNetY-4G,Swin-Tiny和ResNet50对比,精度分别提高0.3%,1.4%和0.5%.
文摘遥感图像的道路分割任务是遥感应用领域的一个研究热点,一直受到广泛的关注。由于遥感图像天然具备背景复杂、目标密集等特性,全局语义信息的构建对于准确提取遥感图像中道路是至关重要的。因此,基于Transformer模型进行优化,提出了基于空间可分离注意力的跨尺度令牌嵌入Transformer遥感道路提取模型Cross-RoadFormer。具体而言,针对图像中道路尺度不统一的问题,设计了跨尺度编码层,将不同尺度的特征编码作为一个令牌嵌入整体,作为Transformer的输入,解决了Transformer跨尺度交互的问题;此外,提出了一种空间可分离注意力,其中,局部分组注意力获取细粒度、短距离信息,全局采样注意力捕获长距离、全局上下文信息,在保证道路提取准确度的前提下,降低了模型的计算量。在Massachusetts数据集和DeepGlobe数据集上的实验表明,提出的Cross-RoadFormer都实现了更高的IoU(intersection over union),分别为68.40%和58.04%,展现了该方法的优越性。