In the engineering database system, multiple versions of a design including engineering drawings should be managed efficiently. The paper proposes an efficient spatial data structure, that is an expansion of the R tre...In the engineering database system, multiple versions of a design including engineering drawings should be managed efficiently. The paper proposes an efficient spatial data structure, that is an expansion of the R tree and HR tree, for version management of engineering drawings. A novel mechanism to manage the difference between drawings is introduced to the HR tree to eliminate redundant duplications and to reduce the amount of storage required for the data structure. Data management mechanism and structural properties of our data structure called the MVR + tree are described.展开更多
针对交通路口图像复杂,小目标难测且目标之间易遮挡以及天气和光照变化引发的颜色失真、噪声和模糊等问题,提出一种基于YOLOv9(You Only Look Once version 9)的交通路口图像的多目标检测算法ITD-YOLOv9(Intersection Target Detection-...针对交通路口图像复杂,小目标难测且目标之间易遮挡以及天气和光照变化引发的颜色失真、噪声和模糊等问题,提出一种基于YOLOv9(You Only Look Once version 9)的交通路口图像的多目标检测算法ITD-YOLOv9(Intersection Target Detection-YOLOv9)。首先,设计CoT-CAFRNet(Chain-of-Thought prompted Content-Aware Feature Reassembly Network)图像增强网络,以提升图像质量,并优化输入特征;其次,加入通道自适应特征融合(iCAFF)模块,以增强小目标及重叠遮挡目标的提取能力;再次,提出特征融合金字塔结构BiHS-FPN(Bi-directional High-level Screening Feature Pyramid Network),以增强多尺度特征的融合能力;最后,设计IF-MPDIoU(Inner-Focaler-Minimum Point Distance based Intersection over Union)损失函数,以通过调整变量因子,聚焦关键样本,并增强泛化能力。实验结果表明,在自制数据集和SODA10M数据集上,ITD-YOLOv9算法的检测精度分别为83.8%和56.3%,检测帧率分别为64.8 frame/s和57.4 frame/s。与YOLOv9算法相比,ITD-YOLOv9算法的检测精度分别提升了3.9和2.7个百分点。可见,所提算法有效实现了交通路口的多目标检测。展开更多
研究多传感器图像融合在隐藏信息检测中的应用,通过异构数据互补与时空对齐机制,构建跨模态特征增强与动态检测的完整技术体系。采用精密时间协议版本2(Precision Time Protocol version 2,PTPv2)实现多源传感器微秒级同步,结合改进型...研究多传感器图像融合在隐藏信息检测中的应用,通过异构数据互补与时空对齐机制,构建跨模态特征增强与动态检测的完整技术体系。采用精密时间协议版本2(Precision Time Protocol version 2,PTPv2)实现多源传感器微秒级同步,结合改进型随机抽样一致性(RANdom SAmple Consensus,RANSAC)算法完成点云与光学数据的空间配准,并引入门控跨模态注意力机制优化可见光、红外及毫米波特征的动态融合。实验结果表明,多传感器图像融合技术在复杂检测场景中可有效增强隐藏信息的时空异常特征突显能力。展开更多
The search operation of spatial data was a principal operation in existent spatial database management system, but the update operation of spatial data such as tracking are occurring frequently in the spatial database...The search operation of spatial data was a principal operation in existent spatial database management system, but the update operation of spatial data such as tracking are occurring frequently in the spatial database management system recently. So, necessity of concurrency improvement among transactions is increasing. In general database management system, many techniques have been studied to solve concurrency problem of transaction. Among them, multi version algorithm does to minimize interference among transactions. However, to apply existent multi version algorithm to improve concurrency of transaction to spatial database management system, the waste of storage happens because it must store entire version for spatial record even if only aspatial data of spatial record is changed. This paper has proposed the record management techniques to manage separating aspatial data version and spatial data version to decrease waste of storage for record version and improve concurrency among transactions.展开更多
视觉语言预训练(VLP)模型通过对比学习等方法,在多模态任务上表现出了优异的性能。然而现有研究忽视了多视角描述带来的好处,以及语义和语法的重要性。为了解决这一问题,文中提出了多视角对比学习和语义增强多模态预训练(Multi-view lea...视觉语言预训练(VLP)模型通过对比学习等方法,在多模态任务上表现出了优异的性能。然而现有研究忽视了多视角描述带来的好处,以及语义和语法的重要性。为了解决这一问题,文中提出了多视角对比学习和语义增强多模态预训练(Multi-view learning and Semantic Enhancement for Multimodal pre-training,MulSE)模型。MulSE主要分为3个部分:1)在融合编码器模型中,引入带有生成器的多视角对比学习;2)提出了一种新的自监督视觉语言预训练任务——多模态文本重排序;3)增加并探寻最优MLM掩码比例,最大化利用视觉信息的能力。通过改进预训练任务,采取多种最优策略,并通过实验验证MulSE增强了模态内部和模态间的理解能力以及对文本语法和语义的理解能力。预训练仅用4×106的数据量,在图文检索任务中就达到了先前大型数据集的效果,且其在视觉问答和视觉蕴含任务上的评估效果优于先前的理解式VLP模型。展开更多
文摘In the engineering database system, multiple versions of a design including engineering drawings should be managed efficiently. The paper proposes an efficient spatial data structure, that is an expansion of the R tree and HR tree, for version management of engineering drawings. A novel mechanism to manage the difference between drawings is introduced to the HR tree to eliminate redundant duplications and to reduce the amount of storage required for the data structure. Data management mechanism and structural properties of our data structure called the MVR + tree are described.
文摘针对交通路口图像复杂,小目标难测且目标之间易遮挡以及天气和光照变化引发的颜色失真、噪声和模糊等问题,提出一种基于YOLOv9(You Only Look Once version 9)的交通路口图像的多目标检测算法ITD-YOLOv9(Intersection Target Detection-YOLOv9)。首先,设计CoT-CAFRNet(Chain-of-Thought prompted Content-Aware Feature Reassembly Network)图像增强网络,以提升图像质量,并优化输入特征;其次,加入通道自适应特征融合(iCAFF)模块,以增强小目标及重叠遮挡目标的提取能力;再次,提出特征融合金字塔结构BiHS-FPN(Bi-directional High-level Screening Feature Pyramid Network),以增强多尺度特征的融合能力;最后,设计IF-MPDIoU(Inner-Focaler-Minimum Point Distance based Intersection over Union)损失函数,以通过调整变量因子,聚焦关键样本,并增强泛化能力。实验结果表明,在自制数据集和SODA10M数据集上,ITD-YOLOv9算法的检测精度分别为83.8%和56.3%,检测帧率分别为64.8 frame/s和57.4 frame/s。与YOLOv9算法相比,ITD-YOLOv9算法的检测精度分别提升了3.9和2.7个百分点。可见,所提算法有效实现了交通路口的多目标检测。
文摘研究多传感器图像融合在隐藏信息检测中的应用,通过异构数据互补与时空对齐机制,构建跨模态特征增强与动态检测的完整技术体系。采用精密时间协议版本2(Precision Time Protocol version 2,PTPv2)实现多源传感器微秒级同步,结合改进型随机抽样一致性(RANdom SAmple Consensus,RANSAC)算法完成点云与光学数据的空间配准,并引入门控跨模态注意力机制优化可见光、红外及毫米波特征的动态融合。实验结果表明,多传感器图像融合技术在复杂检测场景中可有效增强隐藏信息的时空异常特征突显能力。
基金This work is supported by University IT Research Center ProjectKorea
文摘The search operation of spatial data was a principal operation in existent spatial database management system, but the update operation of spatial data such as tracking are occurring frequently in the spatial database management system recently. So, necessity of concurrency improvement among transactions is increasing. In general database management system, many techniques have been studied to solve concurrency problem of transaction. Among them, multi version algorithm does to minimize interference among transactions. However, to apply existent multi version algorithm to improve concurrency of transaction to spatial database management system, the waste of storage happens because it must store entire version for spatial record even if only aspatial data of spatial record is changed. This paper has proposed the record management techniques to manage separating aspatial data version and spatial data version to decrease waste of storage for record version and improve concurrency among transactions.
文摘视觉语言预训练(VLP)模型通过对比学习等方法,在多模态任务上表现出了优异的性能。然而现有研究忽视了多视角描述带来的好处,以及语义和语法的重要性。为了解决这一问题,文中提出了多视角对比学习和语义增强多模态预训练(Multi-view learning and Semantic Enhancement for Multimodal pre-training,MulSE)模型。MulSE主要分为3个部分:1)在融合编码器模型中,引入带有生成器的多视角对比学习;2)提出了一种新的自监督视觉语言预训练任务——多模态文本重排序;3)增加并探寻最优MLM掩码比例,最大化利用视觉信息的能力。通过改进预训练任务,采取多种最优策略,并通过实验验证MulSE增强了模态内部和模态间的理解能力以及对文本语法和语义的理解能力。预训练仅用4×106的数据量,在图文检索任务中就达到了先前大型数据集的效果,且其在视觉问答和视觉蕴含任务上的评估效果优于先前的理解式VLP模型。
