基于自裁剪异构图的NL2SQL模型

自然语言转换为结构化查询语言（NL2SQL）是语义解析领域的重要任务,其核心为对数据库模式和自然语言问句进行联合学习。现有研究通过将整个数据库模式和自然语言问句联合编码构建异构图,使得异构图中引入大量无用信息,并且忽略了数据库模式中不同信息的重要性。为提高NL2SQL模型的逻辑与执行准确率,提出一种基于自裁剪异构图与相对位置注意力机制的NL2SQL模型（SPRELA）。采用序列到序列的框架,使用ELECTRA预训练语言模型作为骨干网络。引入专家知识,对数据库模式和自然语言问句构建初步异构图。基于自然语言问句对初步异构图进行自裁剪,并使用多头相对位置注意力机制编码自裁剪后的数据库模式与自然语言问句。利用树型解码器和预定义的SQL语法,解码生成SQL语句。在Spider数据集上的实验结果表明,SPRELA模型执行准确率达到71.1%,相比于相同参数量级别的RaSaP模型提升了1.1个百分点,能够更好地将数据库模式与自然语言问句对齐,从而理解自然语言查询中的语义信息。     

3D object segmentation using B-Surface

3D object segmentation is important in computer vision such as target detection in biomedical image analysis. A new method, called B-Surface algorithm, is generated for 3D object segmentation. An improved 3D external force field combined with the normalized GVF is utilized. After the initialization of a surface model near the target, B-Surface starts to deform to locate the boundary of the object. First, it overcomes the difficulty that comes from analyzing 3D volume image slice by slice. And the speed of B-Surface deformation is enhanced since the internal forces are not needed to compute in every iteration deformation step. Next, the normal at every surface point can be calculated easily since B-Surface is a continuous deformable model. And it has the ability to achieve high compression ratio (ratio of data to parameters) by presenting the whole surface with only a relatively small number of control points. Experimental results and analysis are presented in this paper. We can see that the B-Surface algorithm can find the surface of the target efficiently.     

Normalized spectral mixture analysis for monitoring urban composition using ETM+ imagery

With rapid urban growth in recent years, understanding urban biophysical composition and dynamics becomes an important research topic. Remote sensing technologies introduce a potentially scientific basis for examining urban composition and monitoring its changes over time. The vegetation-impervious surface-soil (V-I-S) model, in particular, provides a foundation for describing urban/suburban environments and a basis for further urban analyses including urban growth modeling, environmental impact analysis, and socioeconomic factor estimation. This paper develops a normalized spectral mixture analysis (NSMA) method to examine urban composition in Columbus Ohio using Landsat ETM+ data. In particular, a brightness normalization method is applied to reduce brightness variation. Through this normalization, brightness variability within each V-I-S component is reduced or eliminated, thus allowing a single endmember representing each component. Further, with the normalized image, three endmembers, vegetation, impervious surface, and soil, are chosen to model heterogeneous urban composition using a constrained spectral mixture analysis (SMA) model. The accuracy of impervious surface estimation is assessed and compared with two other existing models. Results indicate that the proposed model is a better alternative to existing models, with a root mean square error (RMSE) of 10.1% for impervious surface estimation in the study area.     

异构信息网络数据上的融合概率图模型

针对异构信息网络中存在多种数据目标类型,以及多种数据连接关系的问题,将多个主题模型和马尔科夫逻辑网相结合,提出了一种融合概率图模型。多个主题模型可以分别描述不同数据目标子空间的主题分布,实现对多种数据目标的预处理。用一阶逻辑子句描述的连接规则构建马尔科夫逻辑网,将每个主题模型中的不同数据目标连接起来。通过使用Gibbs采样,可以对异构网络进行参数学习和推理。在国际通用的异构信息网络DBLP数据集上的实验结果表明,使用融合概率图模型能够更好地表示不同的数据目标和连接关系。实验对比了4种典型的分类方法,多次采样得到的分类结果稳定,对作者、文章和会议取得了较好的分类结果。     

Modeling complex heterogeneous objects with non-manifold heterogeneous cells

This paper presents a new approach to model complex heterogeneous objects with simultaneous geometry intricacies as well as complex material distributions. Different from most of the existing approaches, which utilize manifold B-Rep and the assembly representations, the proposed scheme takes advantage of the non-manifoldcellular representations to model the geometries of the heterogeneous objects. With the aid of the cell adjacency information and attribute based reasoning, complex, smooth and versatile material distributions can be defined upon the intricate geometries. Compared with other similar approaches, the proposed scheme (1) generates heterogeneous object models with higher data consistencies and lower redundancies; (2) naturally avoids unnecessary/repetitive computations and thus improves computation efficiencies; (3) represents versatile material variations/distributions using different heterogeneous feature tree (HFT) structures. The detailed representation, associated algorithms and a prototype software package are presented. Example heterogeneous objects modeled with the proposed approach are provided.     

异构软件构件组装模型设计与实现

研究了异构构件组装方法，根据主流构件模型的差异，设计了异构软件构件组装模型，使构件对外呈现一致的构件视图，屏蔽构件的异构性，有效地利用系统资源，最大限度地为应用系统的生成提供支持。     

基于TVDI的大范围干旱区土壤水分遥感反演模型研究

温度植被干旱指数TVDI(Temperature Vegetation Dryness Index)是一种基于光学与热红外遥感通道数据进行植被覆盖区域表层土壤水分反演的方法。当研究区域较大、地表覆盖格局差异显著时,利用TVDI模型来反演陆表土壤水分,精度通常较低。对Sandholt的TVDI土壤水分反演模型进行了改进:利用云掩膜校正和多天平均温度合成来减少云的影响;同时对研究区域地形起伏、覆盖类型差异的影响进行了消除;对TVDI模型干边的模拟方法进行了改进。最后,使用铝盒采样等方法利用新疆地区观测得到的地面数据来拟合改进后的模型参数,并对2009年5月和8月的土壤水分进行了反演实验。与实测数据的比较分析表明,该模型能基本满足大区域土壤水分反演的要求,改进后的模型对新疆地区的土壤水分估算精度有较显著的提高。     

Holes in the ocean: Filling voids in bathymetric lidar data

The mapping of coral reefs may be efficiently accomplished by the use of airborne laser bathymetry. However, there are often data holes within the bathymetry data which must be filled in order to produce a complete representation of the coral habitat. This study presents a method to fill these data holes through data merging and interpolation. The method first merges ancillary digital sounding data with airborne laser bathymetry data in order to populate data points in all areas but particularly those of data holes. What follows is to generate an elevation surface by spatial interpolation based on the merged data points obtained in the first step. We conduct a case study of the Dry Tortugas National Park in Florida and produced an enhanced digital elevation model in the ocean with this method. Four interpolation techniques, including Kriging, natural neighbor, spline, and inverse distance weighted, are implemented and evaluated on their ability to accurately and realistically represent the shallow-water bathymetry of the study area. The natural neighbor technique is found to be the most effective. Finally, this enhanced digital elevation model is used in conjunction with Ikonos imagery to produce a complete, three-dimensional visualization of the study area.     

采用非均匀纹理层的短毛实时绘制

提出一种采用非均匀纹理层来实时绘制真实感短毛的方法,通过混合绘制多层次半透明纹理层来表示物体表面的毛发效果,根据视点位置以及物体表面各部位毛发形态的不同,自适应地采用不同的层数来表示物体各部位的毛发.这种非均匀层数的方式可以在保证绘制质量的情况下尽量减少绘制的面片数,从而加快绘制速度.该方法对中等规模的模型达到了实时的处理速度,并可以有效地表现物体表面的局部性长毛或倒伏毛发等各式毛发形态.     

iMashup: a mashup-based framework for service composition

The Web has undergone a tremendous change from a primarily publication platform towards a participatory and"programmable"platform,where a large number of heterogeneous Web-delivered services(including SOAP and RESTful Web services,RSS and Atom feeds)are emerging.It results in the creation of Web mashup applications with rich user experiences.However,the integration of Web-delivered services is still a challenging issue.It not only requires the developers’tedious eforts in understanding and coordinating heterogeneous service types,but also results in the time-consuming development of user interfaces.In this paper,we propose the iMashup composition framework to facilitate mashup development and deployment.We provide a unified mashup component model for the common representation of heterogeneous Web-delivered service interfaces.The component model specifies necessary properties and behaviors at both business and user interface level.We associate the component model with semantically meaningful tags,so that mashup developers can fast understand the service capabilities.The mashup developers can search and put the proper mashup components into the Web browser based composition environment,and connect them by data flows based on the tag-based semantics.Such an integration manner might prevent some low-level programming eforts and improve the composition efciency.A series of experimental study are conducted to evaluate our framework.