In allusion to the difficulty of integrating data with different models in integrating spatial information, the characteristics of raster structure, vector structure and mixed model were analyzed, and a hierarchical v...In allusion to the difficulty of integrating data with different models in integrating spatial information, the characteristics of raster structure, vector structure and mixed model were analyzed, and a hierarchical vector-raster integrative full feature model was put forward by integrating the advantage of vector and raster model and using the object-oriented method. The data structures of the four basic features, i.e. point, line, surface and solid, were described. An application was analyzed and described, and the characteristics of this model were described. In this model, all objects in the real world are divided into and described as features with hierarchy, and all the data are organized in vector. This model can describe data based on feature, field, network and other models, and avoid the disadvantage of inability to integrate data based on different models and perform spatial analysis on them in spatial information integration.展开更多
A blank panel design algorithm based on feature mapping methods for integral wing skin panels with supercritical airfoil surface is presented.The model of a wing panel is decomposed into features,and features of the p...A blank panel design algorithm based on feature mapping methods for integral wing skin panels with supercritical airfoil surface is presented.The model of a wing panel is decomposed into features,and features of the panel are decomposed into information of location,direction,dimension and Boolean types.Features are mapped into the plane through optimal surface development algorithm.The plane panel is modeled by rebuilding the mapped features.Blanks of shot-peen forming panels are designed to identify the effectiveness of the methods.展开更多
针对遥感地物建筑物图像目标尺度差异大、样本空间分布不均衡、地物边界模糊、场景区域跨度大所导致的分割效果不佳问题,本文提出一种融合动态特征增强高精度遥感建筑物分割算法。首先,构建New_GhostNetV2网络,利用自适应上下文感知卷积...针对遥感地物建筑物图像目标尺度差异大、样本空间分布不均衡、地物边界模糊、场景区域跨度大所导致的分割效果不佳问题,本文提出一种融合动态特征增强高精度遥感建筑物分割算法。首先,构建New_GhostNetV2网络,利用自适应上下文感知卷积,增强算法对样本空间特征的捕捉能力。其次,采用Ghost Convolution结合跳跃连接和特征分支策略设计多层级信息增强模块,增强特征整合。随后引入级联注意力CGA(cascaded group attention),通过组内独立注意力计算,加强模型对多样化地物形态的适应性。最后,通过动态深度特征增强器构造特征融合模块,进一步加强模型捕获能力。在WHU数据集上实验结果表明:改进算法较基线模型F1-Score提高8.57%,mIoU提高12.48%,与其他主流语义分割模型相比,改进DeepLabv3+具有更好的分割精度。展开更多
事件抽取旨在从大量无结构的文本中抽取出结构化的事件信息,然而现有的研究工作存在难以抽取重叠角色,子任务间缺乏交互以及语义特征表达能力不足的问题。针对上述问题提出了一种中文事件抽取模型PACJEE(pre-trained language model and...事件抽取旨在从大量无结构的文本中抽取出结构化的事件信息,然而现有的研究工作存在难以抽取重叠角色,子任务间缺乏交互以及语义特征表达能力不足的问题。针对上述问题提出了一种中文事件抽取模型PACJEE(pre-trained language model and attention mechanism based Chinese joint event extraction)。该模型采用预训练语言模型RoBERTa来提取文本特征,对文本进行事件类型分类,在触发词识别阶段,将提取到的事件类型先验特征与文本特征进行融合,并且使用自注意力机制获取内部特征相关性,在论元角色分类阶段引入卷积神经网络与注意力机制来加强触发词特征的表达能力,通过多层指针标注进行重叠角色的识别。该方法在中文数据集ACE2005和DuEE上进行了实验分析,结果显示,相较于基准方法,在触发词分类上的F1值分别提升1.6和0.5个百分点,在论元角色分类上的F1值分别提升3.3和2.5个百分点,说明该模型能显著提升事件抽取效果,并且在一定程度上提升了对角色重叠事件的识别准确率。展开更多
基金Project (40473029) supported bythe National Natural Science Foundation of China project (04JJ3046) supported bytheNatural Science Foundation of Hunan Province , China
文摘In allusion to the difficulty of integrating data with different models in integrating spatial information, the characteristics of raster structure, vector structure and mixed model were analyzed, and a hierarchical vector-raster integrative full feature model was put forward by integrating the advantage of vector and raster model and using the object-oriented method. The data structures of the four basic features, i.e. point, line, surface and solid, were described. An application was analyzed and described, and the characteristics of this model were described. In this model, all objects in the real world are divided into and described as features with hierarchy, and all the data are organized in vector. This model can describe data based on feature, field, network and other models, and avoid the disadvantage of inability to integrate data based on different models and perform spatial analysis on them in spatial information integration.
文摘A blank panel design algorithm based on feature mapping methods for integral wing skin panels with supercritical airfoil surface is presented.The model of a wing panel is decomposed into features,and features of the panel are decomposed into information of location,direction,dimension and Boolean types.Features are mapped into the plane through optimal surface development algorithm.The plane panel is modeled by rebuilding the mapped features.Blanks of shot-peen forming panels are designed to identify the effectiveness of the methods.
文摘针对遥感地物建筑物图像目标尺度差异大、样本空间分布不均衡、地物边界模糊、场景区域跨度大所导致的分割效果不佳问题,本文提出一种融合动态特征增强高精度遥感建筑物分割算法。首先,构建New_GhostNetV2网络,利用自适应上下文感知卷积,增强算法对样本空间特征的捕捉能力。其次,采用Ghost Convolution结合跳跃连接和特征分支策略设计多层级信息增强模块,增强特征整合。随后引入级联注意力CGA(cascaded group attention),通过组内独立注意力计算,加强模型对多样化地物形态的适应性。最后,通过动态深度特征增强器构造特征融合模块,进一步加强模型捕获能力。在WHU数据集上实验结果表明:改进算法较基线模型F1-Score提高8.57%,mIoU提高12.48%,与其他主流语义分割模型相比,改进DeepLabv3+具有更好的分割精度。
文摘事件抽取旨在从大量无结构的文本中抽取出结构化的事件信息,然而现有的研究工作存在难以抽取重叠角色,子任务间缺乏交互以及语义特征表达能力不足的问题。针对上述问题提出了一种中文事件抽取模型PACJEE(pre-trained language model and attention mechanism based Chinese joint event extraction)。该模型采用预训练语言模型RoBERTa来提取文本特征,对文本进行事件类型分类,在触发词识别阶段,将提取到的事件类型先验特征与文本特征进行融合,并且使用自注意力机制获取内部特征相关性,在论元角色分类阶段引入卷积神经网络与注意力机制来加强触发词特征的表达能力,通过多层指针标注进行重叠角色的识别。该方法在中文数据集ACE2005和DuEE上进行了实验分析,结果显示,相较于基准方法,在触发词分类上的F1值分别提升1.6和0.5个百分点,在论元角色分类上的F1值分别提升3.3和2.5个百分点,说明该模型能显著提升事件抽取效果,并且在一定程度上提升了对角色重叠事件的识别准确率。
