The 500 MHz 5-cell superconducting RF(SRF) cavity was designed aiming to be a candidate cavity for high current accelerators. A copper prototype cavity and a niobium cavity were fabricated at SINAP in 2012. In order t...The 500 MHz 5-cell superconducting RF(SRF) cavity was designed aiming to be a candidate cavity for high current accelerators. A copper prototype cavity and a niobium cavity were fabricated at SINAP in 2012. In order to ensure these cavities get the desired frequency and a good field flatness higher than 98%, frequency control was implemented in the manufacturing process and pre-tuning has been done using a simple pre-tuning frame based on the bead-pull pre-tuning method. Then, TM010-π mode frequency within 5 kHz from the target frequency was achieved and the field flatness reached 98.9% on the copper prototype cavity. Finally, the same procedure was applied to the niobium cavity to obtain a field flatness better than 98% which benefited the cavity performance in the vertical testing.展开更多
腹腔镜手术自动化是智能外科的重要组成部分,其前提是腔镜视野下手术器械与脏器实时精准分割。受术中血液污染、烟雾干扰等复杂因素影响,器械与脏器实时精准分割面临巨大挑战,现有图像分割方法均表现不佳。因此提出一种基于注意力感知...腹腔镜手术自动化是智能外科的重要组成部分,其前提是腔镜视野下手术器械与脏器实时精准分割。受术中血液污染、烟雾干扰等复杂因素影响,器械与脏器实时精准分割面临巨大挑战,现有图像分割方法均表现不佳。因此提出一种基于注意力感知与空间通道的快速分割网络(ASC-Net),以实现腹腔镜图像中器械和脏器快速精准分割。在UNet架构下,设计了注意力感知与空间通道模块,通过跳跃连接将二者嵌入编码与解码模块间,使网络重点关注图像中相似目标间深层语义信息差异,同时多维度学习各目标的多尺度特征。此外,采用预训练微调策略,减小网络计算量。实验结果表明:在EndoVis2018数据集上的平均骰子系数(mDice)、平均重叠度(mIoU)、平均推理时间(mIT)分别为90.64%,86.40%和16.73 ms (60帧/秒),相比于现有最先进方法,mDice与mIoU提升了26%与39%,且mIT降低了56%;在AutoLaparo数据集上的mDice,mIoU和mIT分别为93.72%,89.43%和16.41ms(61帧/秒),同样优于对比方法。该方法在保证分割速度的同时有效提升了分割精度,实现了腹腔镜图像中手术器械和脏器的快速精准分割,将助力腹腔镜手术自动化快速发展。展开更多
针对关系抽取(RE)任务中实体关系语义挖掘困难和预测关系有偏差等问题,提出一种基于掩码提示与门控记忆网络校准(MGMNC)的RE方法。首先,利用提示中的掩码学习实体之间在预训练语言模型(PLM)语义空间中的潜在语义,通过构造掩码注意力权...针对关系抽取(RE)任务中实体关系语义挖掘困难和预测关系有偏差等问题,提出一种基于掩码提示与门控记忆网络校准(MGMNC)的RE方法。首先,利用提示中的掩码学习实体之间在预训练语言模型(PLM)语义空间中的潜在语义,通过构造掩码注意力权重矩阵,将离散的掩码语义空间相互关联;其次,采用门控校准网络将含有实体和关系语义的掩码表示融入句子的全局语义;再次,将它们作为关系提示校准关系信息,随后将句子表示的最终表示映射至相应的关系类别;最后,通过更好地利用提示中掩码,并结合传统微调方法的学习句子全局语义的优势,充分激发PLM的潜力。实验结果表明,所提方法在SemEval(SemEval-2010 Task 8)数据集的F1值达到91.4%,相较于RELA(Relation Extraction with Label Augmentation)生成式方法提高了1.0个百分点;在SciERC(Entities, Relations, and Coreference for Scientific knowledge graph construction)和CLTC(Chinese Literature Text Corpus)数据集上的F1值分别达到91.0%和82.8%。所提方法在上述3个数据集上均明显优于对比方法,验证了所提方法的有效性。相较于基于生成式的方法,所提方法实现了更优的抽取性能。展开更多
文摘The 500 MHz 5-cell superconducting RF(SRF) cavity was designed aiming to be a candidate cavity for high current accelerators. A copper prototype cavity and a niobium cavity were fabricated at SINAP in 2012. In order to ensure these cavities get the desired frequency and a good field flatness higher than 98%, frequency control was implemented in the manufacturing process and pre-tuning has been done using a simple pre-tuning frame based on the bead-pull pre-tuning method. Then, TM010-π mode frequency within 5 kHz from the target frequency was achieved and the field flatness reached 98.9% on the copper prototype cavity. Finally, the same procedure was applied to the niobium cavity to obtain a field flatness better than 98% which benefited the cavity performance in the vertical testing.
文摘腹腔镜手术自动化是智能外科的重要组成部分,其前提是腔镜视野下手术器械与脏器实时精准分割。受术中血液污染、烟雾干扰等复杂因素影响,器械与脏器实时精准分割面临巨大挑战,现有图像分割方法均表现不佳。因此提出一种基于注意力感知与空间通道的快速分割网络(ASC-Net),以实现腹腔镜图像中器械和脏器快速精准分割。在UNet架构下,设计了注意力感知与空间通道模块,通过跳跃连接将二者嵌入编码与解码模块间,使网络重点关注图像中相似目标间深层语义信息差异,同时多维度学习各目标的多尺度特征。此外,采用预训练微调策略,减小网络计算量。实验结果表明:在EndoVis2018数据集上的平均骰子系数(mDice)、平均重叠度(mIoU)、平均推理时间(mIT)分别为90.64%,86.40%和16.73 ms (60帧/秒),相比于现有最先进方法,mDice与mIoU提升了26%与39%,且mIT降低了56%;在AutoLaparo数据集上的mDice,mIoU和mIT分别为93.72%,89.43%和16.41ms(61帧/秒),同样优于对比方法。该方法在保证分割速度的同时有效提升了分割精度,实现了腹腔镜图像中手术器械和脏器的快速精准分割,将助力腹腔镜手术自动化快速发展。
文摘针对关系抽取(RE)任务中实体关系语义挖掘困难和预测关系有偏差等问题,提出一种基于掩码提示与门控记忆网络校准(MGMNC)的RE方法。首先,利用提示中的掩码学习实体之间在预训练语言模型(PLM)语义空间中的潜在语义,通过构造掩码注意力权重矩阵,将离散的掩码语义空间相互关联;其次,采用门控校准网络将含有实体和关系语义的掩码表示融入句子的全局语义;再次,将它们作为关系提示校准关系信息,随后将句子表示的最终表示映射至相应的关系类别;最后,通过更好地利用提示中掩码,并结合传统微调方法的学习句子全局语义的优势,充分激发PLM的潜力。实验结果表明,所提方法在SemEval(SemEval-2010 Task 8)数据集的F1值达到91.4%,相较于RELA(Relation Extraction with Label Augmentation)生成式方法提高了1.0个百分点;在SciERC(Entities, Relations, and Coreference for Scientific knowledge graph construction)和CLTC(Chinese Literature Text Corpus)数据集上的F1值分别达到91.0%和82.8%。所提方法在上述3个数据集上均明显优于对比方法,验证了所提方法的有效性。相较于基于生成式的方法,所提方法实现了更优的抽取性能。