To reduce the computation cost of a combined probabilistic graphical model and a deep neural network in semantic segmentation, the local region condition random field (LRCRF) model is investigated which selectively ap...To reduce the computation cost of a combined probabilistic graphical model and a deep neural network in semantic segmentation, the local region condition random field (LRCRF) model is investigated which selectively applies the condition random field (CRF) to the most active region in the image. The full convolutional network structure is optimized with the ResNet-18 structure and dilated convolution to expand the receptive field. The tracking networks are also improved based on SiameseFC by considering the frame relations in consecutive-frame traffic scene maps. Moreover, the segmentation results of the greyscale input data sets are more stable and effective than using the RGB images for deep neural network feature extraction. The experimental results show that the proposed method takes advantage of the image features directly and achieves good real-time performance and high segmentation accuracy.展开更多
Maxwell equations were originally designed to describe classic electromagnetic phenomena in any type of medium. In particular, to describe electromagnetic phenomena under the quasistatic electric approximation in medi...Maxwell equations were originally designed to describe classic electromagnetic phenomena in any type of medium. In particular, to describe electromagnetic phenomena under the quasistatic electric approximation in media that are electrically inhomogeneous and isotropic, such as for example when there are strong spatial variations of conductivity, the formalism must be adapted according to the problem considered. We review here two approaches to this problem, first a “microscopic” model, where the spatial variations of conductivity and permittivity are explicitly taken into account. In a second “macroscopic” model, these spatial variations are taken on average by using a mean-field formulation of Maxwell equations. Both of these models can describe the electromagnetic behavior of inhomogeneous media. We illustrate this formalism to describe the electric behavior of biological media, such as brain tissue, which is typically very inhomogeneous. We show that the theory predicts that for the typical frequency range of biological phenomena (lower than about 1000 Hz), the inhomogeneous nature of the medium has a determinant influence.展开更多
In recent years, as the enrollment rate of Chinese colleges has increased year by year, the identification of needy undergraduates has become increasingly important. However, the traditional way to identify college st...In recent years, as the enrollment rate of Chinese colleges has increased year by year, the identification of needy undergraduates has become increasingly important. However, the traditional way to identify college students with financial difficulties mainly relies on manual review and collective voting, which easily causes subjectivity and randomness. To alleviate the problem above, this paper establishes an automatic identification model for needy undergraduates based on the 1842 questionnaires collected from undergraduates in WHUT. Firstly, this paper filters the questionnaire preliminary using the local outlier factor algorithm. Secondly, this paper combines mutual information, Spearman rank correlation coefficient and distance correlation coefficient by rank-sum ratio to select features for eliminating noise from irrelevant features. Thirdly, this paper trains filed-aware factor machine model and compares it with other models, such as Logistic Regression, SVM, etc. Eventually, this paper finds that filed-aware factor machine performers much better than other models in the identification of needy undergraduates, and prominent features affecting the identification of needy undergraduates are the year of the family income, cost of living provided parents, etc.展开更多
A method to predict near-field strong ground motions for scenario earthquakes on active faults is proposed. First, macro-source parameters characterizing the entire source area, i.e., global source parameters, includi...A method to predict near-field strong ground motions for scenario earthquakes on active faults is proposed. First, macro-source parameters characterizing the entire source area, i.e., global source parameters, including fault length, fault width, rupture area, average slip on the fault plane, etc., are estimated by seismogeology survey, seismicity and seismic scaling laws. Second, slip distributions characterizing heterogeneity or roughness on the fault plane, i.e., local source parameters, are reproduced/evaluated by the hybrid slip model. Finally, the finite fault source model, developed from both the global and local source parameters, is combined with the stochastically synthetic technique of ground motion using the dynamic comer frequency based on seismology. The proposed method is applied to simulate the acceleration time histories on three base-rock stations during the 1994 Northridge earthquake. Comparisons between the predicted and recorded acceleration time histories show that the method is feasible and practicable.展开更多
在深海匹配场(Matched Field Processing-MFP)声源定位中,拷贝场模型的计算精度和速度对声源的定位效果和定位效率有着直接的影响。针对这一问题,本文数值仿真研究了深海匹配场定位中采用高斯射线束理论作为拷贝场计算模型的可行性。高...在深海匹配场(Matched Field Processing-MFP)声源定位中,拷贝场模型的计算精度和速度对声源的定位效果和定位效率有着直接的影响。针对这一问题,本文数值仿真研究了深海匹配场定位中采用高斯射线束理论作为拷贝场计算模型的可行性。高斯射线束理论具有计算速度快,物理意义清晰,并且适合并行化处理等优点。文中利用不同水文环境、不同频域范围的大量仿真实验来证明该模型的适用性。结果表明,高斯射线束模型在深远海的声源定位中有很好的定位精度,并且在较低频段同样有效,可以作为深远海声源定位的拷贝场计算模型。展开更多
文摘To reduce the computation cost of a combined probabilistic graphical model and a deep neural network in semantic segmentation, the local region condition random field (LRCRF) model is investigated which selectively applies the condition random field (CRF) to the most active region in the image. The full convolutional network structure is optimized with the ResNet-18 structure and dilated convolution to expand the receptive field. The tracking networks are also improved based on SiameseFC by considering the frame relations in consecutive-frame traffic scene maps. Moreover, the segmentation results of the greyscale input data sets are more stable and effective than using the RGB images for deep neural network feature extraction. The experimental results show that the proposed method takes advantage of the image features directly and achieves good real-time performance and high segmentation accuracy.
文摘Maxwell equations were originally designed to describe classic electromagnetic phenomena in any type of medium. In particular, to describe electromagnetic phenomena under the quasistatic electric approximation in media that are electrically inhomogeneous and isotropic, such as for example when there are strong spatial variations of conductivity, the formalism must be adapted according to the problem considered. We review here two approaches to this problem, first a “microscopic” model, where the spatial variations of conductivity and permittivity are explicitly taken into account. In a second “macroscopic” model, these spatial variations are taken on average by using a mean-field formulation of Maxwell equations. Both of these models can describe the electromagnetic behavior of inhomogeneous media. We illustrate this formalism to describe the electric behavior of biological media, such as brain tissue, which is typically very inhomogeneous. We show that the theory predicts that for the typical frequency range of biological phenomena (lower than about 1000 Hz), the inhomogeneous nature of the medium has a determinant influence.
文摘In recent years, as the enrollment rate of Chinese colleges has increased year by year, the identification of needy undergraduates has become increasingly important. However, the traditional way to identify college students with financial difficulties mainly relies on manual review and collective voting, which easily causes subjectivity and randomness. To alleviate the problem above, this paper establishes an automatic identification model for needy undergraduates based on the 1842 questionnaires collected from undergraduates in WHUT. Firstly, this paper filters the questionnaire preliminary using the local outlier factor algorithm. Secondly, this paper combines mutual information, Spearman rank correlation coefficient and distance correlation coefficient by rank-sum ratio to select features for eliminating noise from irrelevant features. Thirdly, this paper trains filed-aware factor machine model and compares it with other models, such as Logistic Regression, SVM, etc. Eventually, this paper finds that filed-aware factor machine performers much better than other models in the identification of needy undergraduates, and prominent features affecting the identification of needy undergraduates are the year of the family income, cost of living provided parents, etc.
基金China Postdoctoral Science Foundation UnderGrant No. 2005037650 Heilongjiang Province PostdoctoralScience Foundation China EarthquakeAdministration’s Tenth"Five Year Plans" Project
文摘A method to predict near-field strong ground motions for scenario earthquakes on active faults is proposed. First, macro-source parameters characterizing the entire source area, i.e., global source parameters, including fault length, fault width, rupture area, average slip on the fault plane, etc., are estimated by seismogeology survey, seismicity and seismic scaling laws. Second, slip distributions characterizing heterogeneity or roughness on the fault plane, i.e., local source parameters, are reproduced/evaluated by the hybrid slip model. Finally, the finite fault source model, developed from both the global and local source parameters, is combined with the stochastically synthetic technique of ground motion using the dynamic comer frequency based on seismology. The proposed method is applied to simulate the acceleration time histories on three base-rock stations during the 1994 Northridge earthquake. Comparisons between the predicted and recorded acceleration time histories show that the method is feasible and practicable.
文摘在深海匹配场(Matched Field Processing-MFP)声源定位中,拷贝场模型的计算精度和速度对声源的定位效果和定位效率有着直接的影响。针对这一问题,本文数值仿真研究了深海匹配场定位中采用高斯射线束理论作为拷贝场计算模型的可行性。高斯射线束理论具有计算速度快,物理意义清晰,并且适合并行化处理等优点。文中利用不同水文环境、不同频域范围的大量仿真实验来证明该模型的适用性。结果表明,高斯射线束模型在深远海的声源定位中有很好的定位精度,并且在较低频段同样有效,可以作为深远海声源定位的拷贝场计算模型。
