近年来,Transformer在众多监督式计算机视觉任务中取得了显著进展,然而由于高质量医学标注图像的缺乏,其在半监督图像分割领域的性能仍有待提高。为此,提出了一种基于多尺度和多视图Transformer的半监督医学图像分割框架:MSMVT(multi-sc...近年来,Transformer在众多监督式计算机视觉任务中取得了显著进展,然而由于高质量医学标注图像的缺乏,其在半监督图像分割领域的性能仍有待提高。为此,提出了一种基于多尺度和多视图Transformer的半监督医学图像分割框架:MSMVT(multi-scale and multi-view transformer)。鉴于对比学习在Transformer的预训练中取得的良好效果,设计了一个基于伪标签引导的多尺度原型对比学习模块。该模块利用图像金字塔数据增强技术,为无标签图像生成富有语义信息的多尺度原型表示;通过对比学习,强化了不同尺度原型之间的一致性,从而有效缓解了由标签稀缺性导致的Transformer训练不足的问题。此外,为了增强Transformer模型训练的稳定性,提出了多视图一致性学习策略。通过弱扰动视图,以校正多个强扰动视图。通过最小化不同视图之间的输出差异性,使得模型能够对不同扰动保持多层次的一致性。实验结果表明,当仅采用10%的标注比例时,提出的MSMVT框架在ACDC、LIDC和ISIC三个公共数据集上的DSC图像分割性能指标分别达到了88.93%、84.75%和85.38%,优于现有的半监督医学图像分割方法。展开更多
Solid-state transformers(SSTs)have been widely used in many areas owing to their advantages of high-frequency isolation and high power density.However,high-frequency switching causes severe electromagnetic interferenc...Solid-state transformers(SSTs)have been widely used in many areas owing to their advantages of high-frequency isolation and high power density.However,high-frequency switching causes severe electromagnetic interference(EMI)problems.Particularly,the common-mode(CM)EMI caused by the switching of the dual active bridge(DAB)converter is conducted through the parasitic capacitances in the high-frequency transformer and impacts the system reliability.With the understanding of the CM EMI model in SSTs,CM EMI mitigation methods have been studied.For passive mitigation,the coupled inductor can be integrated with the phase-shift inductor function to reduce CM EMI.For active mitigation,variations in the DAB switching frequency can help reduce the CM EMI peak.An active EMI filter can also be designed to sample and compensate for CM EMI.Using these methods,CM EMI can be reduced in SSTs.展开更多
文摘近年来,Transformer在众多监督式计算机视觉任务中取得了显著进展,然而由于高质量医学标注图像的缺乏,其在半监督图像分割领域的性能仍有待提高。为此,提出了一种基于多尺度和多视图Transformer的半监督医学图像分割框架:MSMVT(multi-scale and multi-view transformer)。鉴于对比学习在Transformer的预训练中取得的良好效果,设计了一个基于伪标签引导的多尺度原型对比学习模块。该模块利用图像金字塔数据增强技术,为无标签图像生成富有语义信息的多尺度原型表示;通过对比学习,强化了不同尺度原型之间的一致性,从而有效缓解了由标签稀缺性导致的Transformer训练不足的问题。此外,为了增强Transformer模型训练的稳定性,提出了多视图一致性学习策略。通过弱扰动视图,以校正多个强扰动视图。通过最小化不同视图之间的输出差异性,使得模型能够对不同扰动保持多层次的一致性。实验结果表明,当仅采用10%的标注比例时,提出的MSMVT框架在ACDC、LIDC和ISIC三个公共数据集上的DSC图像分割性能指标分别达到了88.93%、84.75%和85.38%,优于现有的半监督医学图像分割方法。
文摘Solid-state transformers(SSTs)have been widely used in many areas owing to their advantages of high-frequency isolation and high power density.However,high-frequency switching causes severe electromagnetic interference(EMI)problems.Particularly,the common-mode(CM)EMI caused by the switching of the dual active bridge(DAB)converter is conducted through the parasitic capacitances in the high-frequency transformer and impacts the system reliability.With the understanding of the CM EMI model in SSTs,CM EMI mitigation methods have been studied.For passive mitigation,the coupled inductor can be integrated with the phase-shift inductor function to reduce CM EMI.For active mitigation,variations in the DAB switching frequency can help reduce the CM EMI peak.An active EMI filter can also be designed to sample and compensate for CM EMI.Using these methods,CM EMI can be reduced in SSTs.