A measuring system for high-speed hydrogen ice pellet was introduced in this paper and the photographing of flying pellet taken therein. With the system, a pellet (minimum size of φ 0.3 mm) velocity (ranging from 50 ...A measuring system for high-speed hydrogen ice pellet was introduced in this paper and the photographing of flying pellet taken therein. With the system, a pellet (minimum size of φ 0.3 mm) velocity (ranging from 50 m/s to 2500 m/s) can be measured in the HL-1M tokamak fueling experiments. By analyzing photographs and the conditions of frozen pellets (including gas supply, gas replenishment, temperature controlling etc), the pellet-freezing technology is summarized in the paper.展开更多
The roles of laser-induced defects and native defects in multilayer mirrors under multi-shot irradiation condition are investigated. The HfO2/SiO2 dielectric mirrors are deposited by electron beam evaporation （EBE）....The roles of laser-induced defects and native defects in multilayer mirrors under multi-shot irradiation condition are investigated. The HfO2/SiO2 dielectric mirrors are deposited by electron beam evaporation （EBE）. Laser damage testing is carried out on both the 1-on-1 and S-on-1 regimes using 355-nm pulsed laser at a duration of 8 ns. It is found that the single-shot laser-induced damage threshold （LIDT） is much higher than the multi-shot LIDT. In the multi-shot mode, the main factor influencing LIDT is the accumulation of irreversible laser-induced defects and native defects. The surface morphologies of the samples are observed by optical microscopy. Moreover, the number of laser-induced defects affects the damage probability of the samples. A correlative model based on critical conduction band （CB） electron density （ED） is presented to simulate the multi-shot damage behavior.展开更多
A multi-shot target assembly and automatic alignment procedure for laser–plasma proton acceleration at high repetition rate are introduced.The assembly is based on a multi-target rotating wheel capable of hosting mor...A multi-shot target assembly and automatic alignment procedure for laser–plasma proton acceleration at high repetition rate are introduced.The assembly is based on a multi-target rotating wheel capable of hosting more than 5000 targets,mounted on a 3D motorized stage to allow rapid replenishment and alignment of the target material between laser irradiations.The automatic alignment procedure consists of a detailed mapping of the impact positions at the target surface prior to the irradiation that ensures stable operation of the target,which alongside the purpose-built design of the target wheel,enables operation at rates up to 10 Hz.Stable and continuous laser-driven proton acceleration at 10 Hz is demonstrated,with observed cut-off energy stability about 15%.展开更多
本文针对一个新的、具有挑战性的问题,即实现源域、中间域到单个目标域的知识转移,其中目标域的每个类别仅有少量标记样本.此种情况下的知识转移过程面临以下两个困难:一是目标数据极其稀缺,从而导致没有足够的目标域特征分布;二是现有...本文针对一个新的、具有挑战性的问题,即实现源域、中间域到单个目标域的知识转移,其中目标域的每个类别仅有少量标记样本.此种情况下的知识转移过程面临以下两个困难:一是目标数据极其稀缺,从而导致没有足够的目标域特征分布;二是现有的少样本学习方法无差别地提取每部分特征,从而导致少样本目标检测性能不佳.为解决上述问题,本文提出一种少样本多源域目标检测方法.提出一种新的元优化机制,通过引入的混合域将源域和目标域对齐,用以缓解目标域稀缺特征分布的问题.具体而言,首先利用图像级混合生成混合图像,和相应的标签共同构成第一个混合域;然后通过双通道注意力机制生成细粒度特征,再利用特征级混合生成特征级混合特征,和相应的标签共同构成第二个混合域;最后通过区域建议网络和感兴趣区域网络生成感兴趣区域特征,再利用感兴趣区域特征级混合生成ROI(Region Of Interest)级混合ROI特征,和相应的标签共同构成第三个混合域.生成的三个混合域共同用于计算损失函数,完成元优化过程.提出一种包含卷积层和特征校准的双通道注意力机制,用以学习更具判别性的深度特征表征,其中卷积层用于防止关键空间信息的丢失,特征校准用于选择性地增强重要特征并削弱非重要特征.具体而言,首先利用卷积层子模块生成粗粒度特征表示;其次,利用特征校准子模块根据特征间的相关性建立注意力权重,并将这些注意力权重与原始特征进行整合,从而有选择地强化重要区域,同时抑制不重要区域.COCO数据集和PASCAL-VOC数据集的大量实验结果证明了本文提出的跨域元优化和双通道注意力结合的少样本多源域目标检测方法的有效性和鲁棒性.在检测效果上超越了同领域内其他方法,同时在不同数据集上保持了良好的泛化性能,此外模型的参数量在同领域内相比其他方法有显著优势.展开更多
针对红外与可见光图像融合中的颜色失真和热目标细节丢失问题,提出基于融合曲线的零样本红外与可见光图像融合方法(Zero-Shot Infrared and Visible Image Fusion Based on Fusion Curve,ZSFuCu).首先,将融合任务转化为基于深度网络的...针对红外与可见光图像融合中的颜色失真和热目标细节丢失问题,提出基于融合曲线的零样本红外与可见光图像融合方法(Zero-Shot Infrared and Visible Image Fusion Based on Fusion Curve,ZSFuCu).首先,将融合任务转化为基于深度网络的图像特定曲线估计过程,通过像素级非线性映射实现热目标纹理的增强与色彩特征的保留.然后,设计多维度视觉感知损失函数,从对比度增强、颜色保持及空间连续性三个维度构建约束机制,协同优化融合图像的高频信息与色彩分布,保留结构特征和关键信息.最后,采用零样本训练策略,仅需单个红外与可见光图像对即可完成参数的自适应优化,具备在不同照明条件下融合的强鲁棒性.实验表明,ZSFuCu在目标突出性、细节丰富度及颜色自然度方面具有显著优势,兼具有效性与实用性.展开更多
在细粒度图像分类中,现有的小样本学习算法未能充分结合通道和空间信息提取细粒度图像的判别性特征,导致仅依靠单一类型的特征不足以准确捕捉细粒度对象的类间差异.针对这一难题,提出了一种基于通道先验感知的多尺度细化网络,旨在有效...在细粒度图像分类中,现有的小样本学习算法未能充分结合通道和空间信息提取细粒度图像的判别性特征,导致仅依靠单一类型的特征不足以准确捕捉细粒度对象的类间差异.针对这一难题,提出了一种基于通道先验感知的多尺度细化网络,旨在有效融合通道信息和空间信息,提升小样本细粒度图像分类的性能.通道先验感知模块实现了通道维度上注意力权重的动态分配,从而高效地捕捉通道先验信息;多尺度特征聚合过程充分利用细粒度图像中丰富的上下文信息,获取丰富的空间和边界细节特征;最后,特征细化模块对上述提取的通道和空间维度信息进行优化,实现了对关键区域的动态选择和强调,进而融合形成更精细、更具代表性的混合特征表示.所提算法在以Conv-4作为骨干网络时,在Stanford Cars、Stanford Dogs和CUB-200-2011三个细粒度数据集上的实验分类性能显著提升.在5 way 1 shot分类任务中,三个数据集的准确率分别达到了79.95%、66.97%和81.91%;在5 way 5 shot分类任务中,准确率则分别为93.42%、82.48%和93.19%.展开更多
文摘A measuring system for high-speed hydrogen ice pellet was introduced in this paper and the photographing of flying pellet taken therein. With the system, a pellet (minimum size of φ 0.3 mm) velocity (ranging from 50 m/s to 2500 m/s) can be measured in the HL-1M tokamak fueling experiments. By analyzing photographs and the conditions of frozen pellets (including gas supply, gas replenishment, temperature controlling etc), the pellet-freezing technology is summarized in the paper.
文摘The roles of laser-induced defects and native defects in multilayer mirrors under multi-shot irradiation condition are investigated. The HfO2/SiO2 dielectric mirrors are deposited by electron beam evaporation （EBE）. Laser damage testing is carried out on both the 1-on-1 and S-on-1 regimes using 355-nm pulsed laser at a duration of 8 ns. It is found that the single-shot laser-induced damage threshold （LIDT） is much higher than the multi-shot LIDT. In the multi-shot mode, the main factor influencing LIDT is the accumulation of irreversible laser-induced defects and native defects. The surface morphologies of the samples are observed by optical microscopy. Moreover, the number of laser-induced defects affects the damage probability of the samples. A correlative model based on critical conduction band （CB） electron density （ED） is presented to simulate the multi-shot damage behavior.
基金This work was supportedby the Spanish Ministerio de Ciencia,Innovacion y Universidades under grants RTI2018-101578-B-C21,RTI2018-101578-B-C22,FPI predoctorals BES-2017-08917 and PRE2019-090730Unidad de Excelencia Maria de Maetzu under project MdM-2016-0692-17-2+2 种基金the Xuntade Galicia grants GRC ED431C 2017/54 and ED431F2023/21 and a grant of the program Grupos de investigacion consolidados(CIAICO/2022/008)financed by Generalitat Valenciana.Action co-financed by the European Union through the Programa Operativo del Fondo Europeo de Desarrollo Regional(FEDER)of the Comunitat Valenciana 2014-2020(IDIFEDER/2021/002)This work was supported by‘la Caixa'Foundation(ID 100010434)(fellowship code LCF/BQ/PI20/11760027)grant RYC2021-032654-I funded by MICIU/AEI/10.13039/501100011033 and by‘European Union Next Generation EU'.
文摘A multi-shot target assembly and automatic alignment procedure for laser–plasma proton acceleration at high repetition rate are introduced.The assembly is based on a multi-target rotating wheel capable of hosting more than 5000 targets,mounted on a 3D motorized stage to allow rapid replenishment and alignment of the target material between laser irradiations.The automatic alignment procedure consists of a detailed mapping of the impact positions at the target surface prior to the irradiation that ensures stable operation of the target,which alongside the purpose-built design of the target wheel,enables operation at rates up to 10 Hz.Stable and continuous laser-driven proton acceleration at 10 Hz is demonstrated,with observed cut-off energy stability about 15%.
文摘本文针对一个新的、具有挑战性的问题,即实现源域、中间域到单个目标域的知识转移,其中目标域的每个类别仅有少量标记样本.此种情况下的知识转移过程面临以下两个困难:一是目标数据极其稀缺,从而导致没有足够的目标域特征分布;二是现有的少样本学习方法无差别地提取每部分特征,从而导致少样本目标检测性能不佳.为解决上述问题,本文提出一种少样本多源域目标检测方法.提出一种新的元优化机制,通过引入的混合域将源域和目标域对齐,用以缓解目标域稀缺特征分布的问题.具体而言,首先利用图像级混合生成混合图像,和相应的标签共同构成第一个混合域;然后通过双通道注意力机制生成细粒度特征,再利用特征级混合生成特征级混合特征,和相应的标签共同构成第二个混合域;最后通过区域建议网络和感兴趣区域网络生成感兴趣区域特征,再利用感兴趣区域特征级混合生成ROI(Region Of Interest)级混合ROI特征,和相应的标签共同构成第三个混合域.生成的三个混合域共同用于计算损失函数,完成元优化过程.提出一种包含卷积层和特征校准的双通道注意力机制,用以学习更具判别性的深度特征表征,其中卷积层用于防止关键空间信息的丢失,特征校准用于选择性地增强重要特征并削弱非重要特征.具体而言,首先利用卷积层子模块生成粗粒度特征表示;其次,利用特征校准子模块根据特征间的相关性建立注意力权重,并将这些注意力权重与原始特征进行整合,从而有选择地强化重要区域,同时抑制不重要区域.COCO数据集和PASCAL-VOC数据集的大量实验结果证明了本文提出的跨域元优化和双通道注意力结合的少样本多源域目标检测方法的有效性和鲁棒性.在检测效果上超越了同领域内其他方法,同时在不同数据集上保持了良好的泛化性能,此外模型的参数量在同领域内相比其他方法有显著优势.
文摘针对红外与可见光图像融合中的颜色失真和热目标细节丢失问题,提出基于融合曲线的零样本红外与可见光图像融合方法(Zero-Shot Infrared and Visible Image Fusion Based on Fusion Curve,ZSFuCu).首先,将融合任务转化为基于深度网络的图像特定曲线估计过程,通过像素级非线性映射实现热目标纹理的增强与色彩特征的保留.然后,设计多维度视觉感知损失函数,从对比度增强、颜色保持及空间连续性三个维度构建约束机制,协同优化融合图像的高频信息与色彩分布,保留结构特征和关键信息.最后,采用零样本训练策略,仅需单个红外与可见光图像对即可完成参数的自适应优化,具备在不同照明条件下融合的强鲁棒性.实验表明,ZSFuCu在目标突出性、细节丰富度及颜色自然度方面具有显著优势,兼具有效性与实用性.
文摘在细粒度图像分类中,现有的小样本学习算法未能充分结合通道和空间信息提取细粒度图像的判别性特征,导致仅依靠单一类型的特征不足以准确捕捉细粒度对象的类间差异.针对这一难题,提出了一种基于通道先验感知的多尺度细化网络,旨在有效融合通道信息和空间信息,提升小样本细粒度图像分类的性能.通道先验感知模块实现了通道维度上注意力权重的动态分配,从而高效地捕捉通道先验信息;多尺度特征聚合过程充分利用细粒度图像中丰富的上下文信息,获取丰富的空间和边界细节特征;最后,特征细化模块对上述提取的通道和空间维度信息进行优化,实现了对关键区域的动态选择和强调,进而融合形成更精细、更具代表性的混合特征表示.所提算法在以Conv-4作为骨干网络时,在Stanford Cars、Stanford Dogs和CUB-200-2011三个细粒度数据集上的实验分类性能显著提升.在5 way 1 shot分类任务中,三个数据集的准确率分别达到了79.95%、66.97%和81.91%;在5 way 5 shot分类任务中,准确率则分别为93.42%、82.48%和93.19%.
