For TA15 titanium alloy, slip is the dominant plastic deformation mechanism because of relatively high Al content. In order to reveal the grain-scale stress field and geometrically necessary dislocation（GND） density...For TA15 titanium alloy, slip is the dominant plastic deformation mechanism because of relatively high Al content. In order to reveal the grain-scale stress field and geometrically necessary dislocation（GND） density distribution around the slip traces and phase boundaries where the slip lines are blocked due to Burgers orientation relationship（OR） missing. We experimentally investigated tensile deformation on TA15 titanium alloy up to 2.0% strain at room temperature. The slip traces were observed and identified using high resolution scanning electron microscopy（SEM） and electron backscatter diffraction（EBSD） measurements. The grain-scale stress fields around the slip traces and phase boundaries were calculated by the cross-correlationbased method. Based on strain gradient theories, the density of GND was calculated and analyzed. The results indicate that the grain-scale stress is significantly concentrated at phase/grain boundaries and slip traces. Although there is an obvious GND accumulation in the vicinity of phase and subgrain boundaries, no GND density accumulation appears near the slip traces.展开更多
在细粒度图像检索领域,现有研究成果主要集中于采用深层网络实现判别特征提取与精准定位,忽略了浅层特征信息的重要性,且无法消除背景中的复杂噪声干扰,限制了检索性能的提升。有鉴于此,提出了一种基于多层次特征提取的细粒度图像哈希...在细粒度图像检索领域,现有研究成果主要集中于采用深层网络实现判别特征提取与精准定位,忽略了浅层特征信息的重要性,且无法消除背景中的复杂噪声干扰,限制了检索性能的提升。有鉴于此,提出了一种基于多层次特征提取的细粒度图像哈希检索方法(Fine-grained Deep Hashing image retrieval method based on Multi-level Feature Extraction, FDH-MFE)。该方法主要关注不同层次间特征的关联性,并增强了局部特征的提取能力。首先,提出了一个特征提取模块,旨在从网络的不同阶段提取细粒度特征,并通过图神经网络揭示其潜在的长距离依赖关系,为后续阶段提供更全面和精细的特征表示。其次,设计了一种代理损失算法,使得哈希码分布更加均匀,从而提升细粒度特征的区分能力。最后,通过设计背景抑制算法并结合三元组损失,增强了模型拟合全局分布的能力,使得所提出的方法在细粒度图像检索任务中表现出色。实验结果表明:该方法在4个公开数据集上的平均检索精度相较于次先进方法分别提高了15.03%、10.94%、9.98%和9.78%。展开更多
Our understanding of grain-level bursts of plasticity in polycrystals remains limited by current techniques.By employing a modified Synchrotron transmission X-ray Laue diffraction method(beam size larger than the grai...Our understanding of grain-level bursts of plasticity in polycrystals remains limited by current techniques.By employing a modified Synchrotron transmission X-ray Laue diffraction method(beam size larger than the grain size),we tracked grain rotations for the first 1%of tensile strain,in 4400 time steps.We indexed 33 grains and quantified the magnitude and frequency of intermittent bursts of grain rotation.We interpret these events in terms of bursts of plastic deformation.The events are highly coordinated amongst nearby grains,and their frequency and magnitude,as well as the number of grains participating,peaked at around the onset of full plasticity.At this point,7 out of the 10 indexed grains with orientations favorable for twinning showed significant drops in diffracted intensity(a mean value of 8%),due to twin induced re-orientation.For other orientations,20 out of 23 grains displayed bursts attributable to lattice dislocation glide(interpreted in terms of basal and prismatic <α> slip).The mean value of the magnitude of these bursts is∼0.08°,implying accumulated shear strains of the order of 3×10^(-3).These bursts,in many cases,were due to the activation of more than a single slip/twin system within the grain,and co-ordination amongst neighboring grains also involved collaboration between slip and twinning events.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
基金Funded by National Natural Science Foundation of China(No.51401226)
文摘For TA15 titanium alloy, slip is the dominant plastic deformation mechanism because of relatively high Al content. In order to reveal the grain-scale stress field and geometrically necessary dislocation（GND） density distribution around the slip traces and phase boundaries where the slip lines are blocked due to Burgers orientation relationship（OR） missing. We experimentally investigated tensile deformation on TA15 titanium alloy up to 2.0% strain at room temperature. The slip traces were observed and identified using high resolution scanning electron microscopy（SEM） and electron backscatter diffraction（EBSD） measurements. The grain-scale stress fields around the slip traces and phase boundaries were calculated by the cross-correlationbased method. Based on strain gradient theories, the density of GND was calculated and analyzed. The results indicate that the grain-scale stress is significantly concentrated at phase/grain boundaries and slip traces. Although there is an obvious GND accumulation in the vicinity of phase and subgrain boundaries, no GND density accumulation appears near the slip traces.
文摘在细粒度图像检索领域,现有研究成果主要集中于采用深层网络实现判别特征提取与精准定位,忽略了浅层特征信息的重要性,且无法消除背景中的复杂噪声干扰,限制了检索性能的提升。有鉴于此,提出了一种基于多层次特征提取的细粒度图像哈希检索方法(Fine-grained Deep Hashing image retrieval method based on Multi-level Feature Extraction, FDH-MFE)。该方法主要关注不同层次间特征的关联性,并增强了局部特征的提取能力。首先,提出了一个特征提取模块,旨在从网络的不同阶段提取细粒度特征,并通过图神经网络揭示其潜在的长距离依赖关系,为后续阶段提供更全面和精细的特征表示。其次,设计了一种代理损失算法,使得哈希码分布更加均匀,从而提升细粒度特征的区分能力。最后,通过设计背景抑制算法并结合三元组损失,增强了模型拟合全局分布的能力,使得所提出的方法在细粒度图像检索任务中表现出色。实验结果表明:该方法在4个公开数据集上的平均检索精度相较于次先进方法分别提高了15.03%、10.94%、9.98%和9.78%。
基金the Australian Research Council through the Discovery Grant DP200100727 and Laureate Fellowship FL210100147。
文摘Our understanding of grain-level bursts of plasticity in polycrystals remains limited by current techniques.By employing a modified Synchrotron transmission X-ray Laue diffraction method(beam size larger than the grain size),we tracked grain rotations for the first 1%of tensile strain,in 4400 time steps.We indexed 33 grains and quantified the magnitude and frequency of intermittent bursts of grain rotation.We interpret these events in terms of bursts of plastic deformation.The events are highly coordinated amongst nearby grains,and their frequency and magnitude,as well as the number of grains participating,peaked at around the onset of full plasticity.At this point,7 out of the 10 indexed grains with orientations favorable for twinning showed significant drops in diffracted intensity(a mean value of 8%),due to twin induced re-orientation.For other orientations,20 out of 23 grains displayed bursts attributable to lattice dislocation glide(interpreted in terms of basal and prismatic <α> slip).The mean value of the magnitude of these bursts is∼0.08°,implying accumulated shear strains of the order of 3×10^(-3).These bursts,in many cases,were due to the activation of more than a single slip/twin system within the grain,and co-ordination amongst neighboring grains also involved collaboration between slip and twinning events.
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%.
