摘要
高维场景信息的时序风险感知任务,常因数据规模庞大和结构复杂而遭遇计算瓶颈。为此,提出基于双重注意力的高维场景信息时序风险感知方法。首先,通过交替使用空间窗口注意力和通道组注意力,平衡计算开销与建模能力;然后,以城市内涝风险感知为例,利用开源地图Open‑StreetMap获取北京市13种地理特征数据,生成北京市高维内涝风险数据集,基于点、线和面3种不同的类型建立高维场景信息与城市内涝风险的关联映射;最后,试验结果表明,该方法在城市内涝风险感知任务中优于基线方法。
Temporal risk perception of high‑dimensional scene information usually presents computational challenges due to the large scale and complex structure of data.To address these challenges,a temporal risk perception method of high‑dimensional scene information based on dual attention is proposed.Firstly,the capability of computational overhead and modeling is balanced by alternating between spatial window attention and channel group attention.Then,taking urban waterlogging risk perception as an example,the open source map OpenStreetMap is utilized to obtain 13 types of geographic feature data related to Beijing,and the high‑dimensional waterlogging risk dataset for the city is generated.Correlation mapping between high‑dimensional scene information and urban waterlogging risk are established based on three types of elements:point,line,and surface.Finally,the experimental result shows that,the proposed method outperforms all baseline methods in urban waterlogging risk perception task.
作者
王欣芝
李振南
徐健
骆祥峰
谢少荣
WANG Xinzhi;LI Zhennan;XU Jian;LUO Xiangfeng;XIE Shaorong(School of Computer Engineering and Science,Shanghai University,Shanghai 200444,China;Unit 32023 of PLA,Shenyang 110000,China)
出处
《指挥信息系统与技术》
2025年第4期61-67,共7页
Command Information System and Technology
基金
国家基金委青年科学基金(72204155)
上海市科委自然科学基金(23ZR1423100)资助项目。
