针对现有的知识库关系检测任务对于一些不可见关系无法做到准确的向量表示而出现词汇溢出的问题,提出了基于对抗学习和全局知识信息的关系检测模型。该模型使用对抗学习对知识库关系表示模型进行特征强化,使用TransH(translating on hyp...针对现有的知识库关系检测任务对于一些不可见关系无法做到准确的向量表示而出现词汇溢出的问题,提出了基于对抗学习和全局知识信息的关系检测模型。该模型使用对抗学习对知识库关系表示模型进行特征强化,使用TransH(translating on hyperplanes)模型提取全局知识信息,同时通过联合训练,将全局知识信息融合进关系表示模型中,进一步提升关系模型的表示能力。实验结果表明,提出的融合模型对于关系检测效果有一定的提升,并且缓解了词汇溢出的问题。展开更多
From the viewpoint of Graph Theory this paper builds a town road network graph of regional scale, and proposes numerical vertex degree (Di), rank vertex degree (Dr) and population size vertex degree (Dp) on the base o...From the viewpoint of Graph Theory this paper builds a town road network graph of regional scale, and proposes numerical vertex degree (Di), rank vertex degree (Dr) and population size vertex degree (Dp) on the base of vertex degree (D). Then the indicators of Di, Dr, Dp and mathematical statistics methods are applied to investigating scale-free structure of town road networks in the southern Jiangsu Province. The results show that the distribution of Di does not exhibit scale-free properties, but Dr and Dp do. Additionally the correlation coefficient between Dp and Di is only 0.569, but the spatial correlation between Dp and Dr is very evident on the base of correlation analysis and spatial analysis of GIS. The mutual mechanism between Dp and Dr spatially represents a "Core-Belt" model of town development of regional scale. The town development model is open and clustered, and beneficial to both economic development and ecological protection. And then we suggest that Suzhou City, Wuxi City, Changzhou City and Wujin City control towns' high-density pattern by conducting centralization and consolidation policies, and properly controlling and planning higher rank roads; whereas Nanjing City, Zhenjiang City, Jintan City and Liyang City must strengthen the development of towns along higher rank roads.展开更多
地表温度(Land Surface Temperature,LST)在陆地—大气能量交换等研究中扮演着重要角色。LST随时间变化迅速,且极轨遥感卫星获取的LST的地方太阳时在像元间存在差异,需进行时间归一化以提高LST遥感产品的应用价值。面向MODIS LST产品,基...地表温度(Land Surface Temperature,LST)在陆地—大气能量交换等研究中扮演着重要角色。LST随时间变化迅速,且极轨遥感卫星获取的LST的地方太阳时在像元间存在差异,需进行时间归一化以提高LST遥感产品的应用价值。面向MODIS LST产品,基于FY-4A高时间分辨率的LST产品,引入地表温度日变化模型(DTC),构建了粗细分辨率转换配准方法,提出了基于日变化信息的LST时间归一化模型(Temporal-effect Normalization Model of land surface temperature Based on Diurnal variation information,BDTNM),探讨了时间窗口、归一化时刻与空值情况对模型的影响。利用张掖地区站点实测数据、模拟数据对INA08_2模型和BDTNM模型归一化结果进行验证和评价,结果表明BDTNM方法比INA08_2模型具有更好的稳定性及鲁棒性,精度提高了0.4~1.0 K,并具有一定的空值插补能力,该方法对其他遥感卫星LST的时间归一化也具有一定的借鉴意义。展开更多
文摘针对现有的知识库关系检测任务对于一些不可见关系无法做到准确的向量表示而出现词汇溢出的问题,提出了基于对抗学习和全局知识信息的关系检测模型。该模型使用对抗学习对知识库关系表示模型进行特征强化,使用TransH(translating on hyperplanes)模型提取全局知识信息,同时通过联合训练,将全局知识信息融合进关系表示模型中,进一步提升关系模型的表示能力。实验结果表明,提出的融合模型对于关系检测效果有一定的提升,并且缓解了词汇溢出的问题。
基金Under the auspices of National Natural Science Foundation of China (No. 40435013, No. 40535026)
文摘From the viewpoint of Graph Theory this paper builds a town road network graph of regional scale, and proposes numerical vertex degree (Di), rank vertex degree (Dr) and population size vertex degree (Dp) on the base of vertex degree (D). Then the indicators of Di, Dr, Dp and mathematical statistics methods are applied to investigating scale-free structure of town road networks in the southern Jiangsu Province. The results show that the distribution of Di does not exhibit scale-free properties, but Dr and Dp do. Additionally the correlation coefficient between Dp and Di is only 0.569, but the spatial correlation between Dp and Dr is very evident on the base of correlation analysis and spatial analysis of GIS. The mutual mechanism between Dp and Dr spatially represents a "Core-Belt" model of town development of regional scale. The town development model is open and clustered, and beneficial to both economic development and ecological protection. And then we suggest that Suzhou City, Wuxi City, Changzhou City and Wujin City control towns' high-density pattern by conducting centralization and consolidation policies, and properly controlling and planning higher rank roads; whereas Nanjing City, Zhenjiang City, Jintan City and Liyang City must strengthen the development of towns along higher rank roads.
文摘地表温度(Land Surface Temperature,LST)在陆地—大气能量交换等研究中扮演着重要角色。LST随时间变化迅速,且极轨遥感卫星获取的LST的地方太阳时在像元间存在差异,需进行时间归一化以提高LST遥感产品的应用价值。面向MODIS LST产品,基于FY-4A高时间分辨率的LST产品,引入地表温度日变化模型(DTC),构建了粗细分辨率转换配准方法,提出了基于日变化信息的LST时间归一化模型(Temporal-effect Normalization Model of land surface temperature Based on Diurnal variation information,BDTNM),探讨了时间窗口、归一化时刻与空值情况对模型的影响。利用张掖地区站点实测数据、模拟数据对INA08_2模型和BDTNM模型归一化结果进行验证和评价,结果表明BDTNM方法比INA08_2模型具有更好的稳定性及鲁棒性,精度提高了0.4~1.0 K,并具有一定的空值插补能力,该方法对其他遥感卫星LST的时间归一化也具有一定的借鉴意义。
