Railway transitions experience differential movements due to differences in track system stiffness,track damping characteristics,foundation type,ballast settlement from fouling and/or degradation,as well as fill and s...Railway transitions experience differential movements due to differences in track system stiffness,track damping characteristics,foundation type,ballast settlement from fouling and/or degradation,as well as fill and subgrade settlement.This differential movement is especially problematic for high speed rail infrastructure as the 'bump' at the transition is accentuated at high speeds.Identification of different factors contributing towards this differential movement,as well as development of design and maintenance strategies to mitigate the problem is imperative for the safe and economical operation of both freight and passenger rail networks.This paper presents the research framework and initial instrumentation details from an ongoing research effort at the University of Illinois at Urbana-Champaign.Three bridge approaches experiencing recurrent geometry problems were instrumented using multidepth deflectometers(MDDs) and strain gages to identify different factors contributing to the development of differential movements.展开更多
CT图像肺结节大小、形状和纹理的多样性,导致肺结节的良恶性诊断十分困难。在三维卷积神经网络的基础上,提出了一种基于多深度残差注意力机制的网络(MDRA-net),用于良恶性肺结节分类。MDRA-net通过在残差分支上使用特征融合及迭代分层...CT图像肺结节大小、形状和纹理的多样性,导致肺结节的良恶性诊断十分困难。在三维卷积神经网络的基础上,提出了一种基于多深度残差注意力机制的网络(MDRA-net),用于良恶性肺结节分类。MDRA-net通过在残差分支上使用特征融合及迭代分层融合的方法,提升了网络对结节位置特征及全局特征的感知能力;此外,结合注意力机制,引入projection and excitation模块,利用空间和通道信息进行校准,进一步提升了网络提取特征的能力。在LUNA16数据集上的实验结果表明,MDRA-net分类模型的肺结节检测准确率达96.52%,灵敏度和特异性分别为93.01%和97.77%,较现有的基于深度学习的肺结节良恶性分类模型有较大提升。展开更多
文摘Railway transitions experience differential movements due to differences in track system stiffness,track damping characteristics,foundation type,ballast settlement from fouling and/or degradation,as well as fill and subgrade settlement.This differential movement is especially problematic for high speed rail infrastructure as the 'bump' at the transition is accentuated at high speeds.Identification of different factors contributing towards this differential movement,as well as development of design and maintenance strategies to mitigate the problem is imperative for the safe and economical operation of both freight and passenger rail networks.This paper presents the research framework and initial instrumentation details from an ongoing research effort at the University of Illinois at Urbana-Champaign.Three bridge approaches experiencing recurrent geometry problems were instrumented using multidepth deflectometers(MDDs) and strain gages to identify different factors contributing to the development of differential movements.
文摘CT图像肺结节大小、形状和纹理的多样性,导致肺结节的良恶性诊断十分困难。在三维卷积神经网络的基础上,提出了一种基于多深度残差注意力机制的网络(MDRA-net),用于良恶性肺结节分类。MDRA-net通过在残差分支上使用特征融合及迭代分层融合的方法,提升了网络对结节位置特征及全局特征的感知能力;此外,结合注意力机制,引入projection and excitation模块,利用空间和通道信息进行校准,进一步提升了网络提取特征的能力。在LUNA16数据集上的实验结果表明,MDRA-net分类模型的肺结节检测准确率达96.52%,灵敏度和特异性分别为93.01%和97.77%,较现有的基于深度学习的肺结节良恶性分类模型有较大提升。
