检索结果-内蒙古大学图书馆

Cascading Multi-Granular Attention for Meticulous Liver Vessel Segmentation

学校读者我要写书评

暂无评论

SSRN

SSRN 2024年

作者： Gao, Zhan Wang, Mingju Zhang, Zhonglei Hu, Jiayue Huang, Dan Xu, Xunliang Zhao, Liang Taihe Hospital Hubei Key Laboratory of Embryonic Stem Cell Research Hubei University of Medicine Hubei China School of Computing and Electronic Information Guangxi University Guangxi China College of Computer Science and Electronic Engineering Hunan University Hunan China

Segmentation of liver vessels is essential to disease diagnosis and surgical planning. However, the segmentation performance of existing models could be more satisfactory due to the lack of problem-specific sophisticated models. In particular, 1) the intrinsic spatial connectivity of vessels needs to be sufficiently captured, and 2) the typical receptive field expansion strategy, such as downscaling, is harmful to vessel segmentation, particularly for tiny vessels. Towards this goal, a high-resolution lateral cascading joint multi-granularity attention network targeting vessel segmentation is proposed. The multi-granular attention module expands the receptive field centered on vascular salience, while the lateral cascading structure maintains the vascular details at high resolution and strengthens the weak cues through multi-stage cascades to achieve fine-grained feature formulation guided by vascular connectivity. Results on the widely used datasets 3D-IRCADb and MSD show that the newly conceived model produces more meticulous segmentation details and improved distinguishability compared to the state-of-the-art models. © 2024, The Authors. All rights reserved.

关键词： Diagnosis

Modelling brain-wide neuronal morphology via rooted Cayley trees

学校读者我要写书评

暂无评论

arXiv 2018年

作者： Lin, Congping Huang, Yuanfei Quan, Tingwei Zhang, Yiwei Center for Mathematical Sciences Huazhong University of Science and Technology Wuhan China Hubei Key Lab of Engineering Modeling and Scientific Computing Huazhong University of Science and Technology Wuhan China Britton Chance Center for Biomedical Photonics Wuhan National Laboratory for Optoelectronics-Huazhong University of Science and Technology Wuhan China MoE Key Laboratory for Biomedical Photonics Collaborative Innovation Center for Biomedical Engineering School of Engineering Sciences Huazhong University of Science and Technology Wuhan China

Neuronal morphology is an essential element for brain activity and function. We take advantage of current availability of brain-wide neuron digital reconstructions of the Pyramidal cells from a mouse brain, and analyze several emergent features of brain-wide neuronal morphology. We observe that axonal trees are self-affine while dendritic trees are self-similar. We also show that tree size appear to be random, independent of the number of dendrites within single neurons. Moreover, we consider inhomogeneous branching model which stochastically generates rooted 3-Cayley trees for the brain-wide neuron topology. Based on estimated order-dependent branching probability from actual axonal and dendritic trees, our inhomogeneous model quantitatively captures a number of topological features including size and shape of both axons and dendrites. This sheds lights on a universal mechanism behind the topological formation of brain-wide axonal and dendritic trees. Copyright © 2018, The Authors. All rights reserved.

关键词： Brain

Cascading Multi-Granular Attention for Meticulous Liver Vessel Segmentation

学校读者我要写书评

暂无评论

SSRN

SSRN 2024年

关键词： Diagnosis

Phase-field modeling of dendritic growth with gas bubbles in the solidification of binary alloys

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Zhan, Chengjie Chai, Zhenhua Sun, Dongke Shi, Baochang Geng, Shaoning Jiang, Ping School of Mathematics and Statistics Huazhong University of Science and Technology Wuhan430074 China Institute of Interdisciplinary Research for Mathematics and Applied Science Huazhong University of Science and Technology Wuhan430074 China Hubei Key Laboratory of Engineering Modeling and Scientific Computing Huazhong University of Science and Technology Wuhan430074 China School of Mechanical Engineering Southeast University Nanjing211189 China The State Key Laboratory of Digital Manufacturing Equipment and Technology School of Mechanical Science and Engineering Huazhong University of Science and Technology Wuhan430074 China

In this work, a phase-field model is developed for the dendritic growth with gas bubbles in the solidification of binary alloys. In this model, a total free energy for the complex gas-liquid-dendrite system is proposed through considering the interactions of gas bubbles, liquid melt and solid dendrites, and it can reduce to the energy for gas-liquid flows in the region far from the solid phase, while degenerate to the energy for thermosolutal dendritic growth when the gas bubble disappears. The governing equations are usually obtained by minimizing the total free energy, but here some modifications are made to improve the capacity of the conservative phase-field equation for gas bubbles and convection-diffusion equation for solute transfer. Additionally, through the asymptotic analysis of the thin-interface limit, the present general phase-field model for alloy solidification can match the corresponding free boundary problem, and it is identical to the commonly used models under a specific choice of model parameters. Furthermore, to describe the fluid flow, the incompressible Navier-Stokes equations are adopted in the entire domain including gas, liquid, and solid regions, where the fluid-structure interaction is considered by a simple diffuse-interface method. To test the present phase-field model, the lattice Boltzmann method is used to study several problems of gas-liquid flows, dendritic growth as well as the solidification in presence of gas bubbles, and a good performance of the present model for such complex problems is observed. © 2024, CC BY-NC-ND.

关键词： Solidification

Text's Source Trustworthiness Detection Based on Cognitive Hash

学校读者我要写书评

暂无评论

Text's Source Trustworthiness Detection Based on Cognitive H...

International Conference on Multimedia Information Networking and Security (MINES)

作者： Donghui Hu Juan Zhang Xuegang Hu Lian Wang Wenjie Miao Zhengfeng Hou School of Computer Science and Information Engineering Hefei University of Technology Hefei China School of Computer Science Wuhan University of China Wuhan Hubei China The Key Laboratory of AeroSpace Information Security and Trust Computing Ministry of Education Wuhan Hubei China

It is important to do research on the source trustworthiness of the massive data under network environment. A detection method of source trustworthiness in text is proposed and designed based on cognitive hash. Based on the HowNet semantic features, this paper designs a text cognitive hash, and proposes a method evaluating feasible hash distance. Experimental results show the effectiveness of the proposed cognitive hash method.

关键词： Feature extraction Robustness Semantics Training Security Data mining Multimedia communication

A diffuse-interface lattice Boltzmann method for the dendritic growth with thermosolutal convection

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Zhan, Chengjie Chai, Zhenhua Shi, Baochang Jiang, Ping Geng, Shaoning Sun, Dongke School of Mathematics and Statistics Huazhong University of Science and Technology Wuhan430074 China Institute of Interdisciplinary Research for Mathematics and Applied Science Huazhong University of Science and Technology Wuhan430074 China Hubei Key Laboratory of Engineering Modeling and Scientific Computing Huazhong University of Science and Technology Wuhan430074 China The State Key Laboratory of Digital Manufacturing Equipment and Technology School of Mechanical Science and Engineering Huazhong University of Science and Technology Wuhan430074 China School of Mechanical Engineering Southeast University Nanjing211189 China

In this work, we proposed a diffuse interface model for the dendritic growth with thermosolutal convection. In this model, the sharp boundary between the fluid and solid dendrite is replaced by a thin but nonzero thickness diffuse interface, which is described by the order parameter governed by the phase-field equation for the dendritic growth. The governing equations for solute and heat transfer are modified such that the previous special treatments for source term can be avoided. To solve the model for the dendritic growth with thermosolutal convection, we also developed a diffuse-interface multi-relaxation-time lattice Boltzmann (LB) method. In this method, the order parameter in the phase-field equation is combined into the force caused by the fluid-solid interaction, and the treatment on the complex fluid-solid interface can be avoided. In addition, four LB models are developed for the phase-field equation, concentration equation, temperature equation and the Navier-Stokes equations in a unified framework. Finally, to test the present diffuse-interface LB method, we performed some simulations of the dendritic growth, and found that the numerical results are in good agreements with some previous works. © 2022, CC BY-NC-ND.

关键词： Navier Stokes equations

OBSERVABLE SETS, POTENTIALS AND SCHRÖDINGER EQUATIONS

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Huang, Shanlin Wang, Gengsheng Wang, Ming School of Mathematics and Statistics Hubei Key Laboratory of Engineering Modeling and Scientific Computing Huazhong University of Science and Technology Wuhan430074 China Center for Applied Mathematics Tianjin University Tianjin300072 China School of Mathematics and Physics China University of Geosciences Wuhan430074 China

We characterize observable sets for 1-dim Schrödinger equations in R: i∂tu = (−∂2x + x2m)u (with m ∈ N:= {0, 1, . . . }). More precisely, we obtain what follows: First, when m = 0, E ⊂ R is an observable set at some time if and only if it is thick, namely, there is γ > 0 and L > 0 so that E \ [x, x + L] ≥ γL for each x ∈ R;Second, when m = 1 (m ≥ 2 resp.), E is an observable set at some time (at any time resp.) if and only if it is weakly thick, namely lim |E T[−x, x]| > 0. x x→+∞ From these, we see how potentials x2m affect the observability (including the geometric structures of observable sets and the minimal observable time). Besides, we obtain several supplemental theorems for the above results, in particular, we find that a half line is an observable set at time T > 0 for the above equation with m = 1 if and only if T > π2 .MSC Codes 93B07, 35J10 Copyright © 2020, The Authors. All rights reserved.

关键词：

Characterizations of stabilizable sets for some parabolic equations in Rn

学校读者我要写书评

暂无评论

arXiv 2020年

We consider the parabolic type equation in Rn: (∂t + H)y(t, x) = 0, (t, x) ∈ (0, ∞) × Rn;y(0, x) ∈ L2(Rn), (0.1) where H can be one of the following operators: (i) a shifted fractional Laplacian;(ii) a shifted Hermite operator;(iii) the Schrödinger operator with some general potentials. We call a subset E ⊂ Rn as a stabilizable set for (0.1), if there is a linear bounded operator K on L2(Rn) so that the semigroup {e−t(H−χEK)}t≥0 is exponentially stable. (Here, χE denotes the characteristic function of E, which is treated as a linear operator on L2(Rn).) This paper presents different geometric characterizations of the stabilizable sets for (0.1) with different H. In particular, when H is a shifted fractional Laplacian, E ⊂ Rn is a stabilizable set for (0.1) if and only if E ⊂ Rn is a thick set, while when H is a shifted Hermite operator, E ⊂ Rn is a stabilizable set for (0.1) if and only if E ⊂ Rn is a set of positive measure. Our results, together with the results on the observable sets for (0.1) obtained in [1, 18, 24, 32], reveal such phenomena: for some H, the class of stabilizable sets contains strictly the class of observable sets, while for some other H, the classes of stabilizable sets and observable sets coincide. Besides, this paper gives some sufficient conditions on the stabilizable sets for (0.1) where H is the Schrödinger operator with some general potentials. Copyright © 2020, The Authors. All rights reserved.

关键词： Partial differential equations

VDCRL: Vulnerability Detection with Supervised Contrastive Code Representation Learning

学校读者我要写书评

暂无评论

SSRN

SSRN 2024年

作者： Lv, Xinghang Fu, Jianming Nie, Yu The Key Laboratory of Aerospace Information Security and Trusted Computing Ministry of Education School of Cyber Science and Engineering Wuhan University Hubei Wuhan430000 China Jingdezhen Ceramic Institute China

Code vulnerability detection is a critical part of ensuring software security, and recent deep learning-based code vulnerability detection methods have proven their effectiveness. However, the currently available methods show high vulnerability detection ability only on the datasets they use, and when applying them to new datasets, the performance drops significantly, that is, their generalization ability is generally very poor, which does not meet the needs of real-world vulnerability detection. To address the above issues, we propose VDCRL, a vulnerability detection framework based on supervised contrastive code representation learning. VDCRL performs input space-based and feature space-based code augmentation methods, generates two augmentations of the original sample, then extracts the source code features and assembly features of each augmentation sample and builds the feature fusion encoder (SAFE) to fuse the two types of features for embedding, finally, it incorporates supervised contrastive learning and the BGRU model for training and detecting vulnerabilities. We train VDCRL on a synthetic dataset and test its performance on two real datasets, the experimental results show that the vulnerability detection performance of VDCRL outperforms the current state-of-the-art vulnerability detectors. © 2024, The Authors. All rights reserved.

关键词： Contrastive Learning