首页 | 本学科首页   官方微博 | 高级检索  
     检索      

基于注意力反向投影网络的天气雷达回波超分辨率重建算法
引用本文:余秋,曾强宇,张福贵,王皓,史朝,李浩然.基于注意力反向投影网络的天气雷达回波超分辨率重建算法[J].气象科技,2023,51(3):319-330.
作者姓名:余秋  曾强宇  张福贵  王皓  史朝  李浩然
作者单位:1 成都信息工程大学电子工程学院,成都 610225; 2 大气探测重点开放实验室,成都 610225
基金项目:国家自然科学基金项目(U20B2061)和四川省科技厅重点项目(2022YFS0541)资助
摘    要:高分辨率的天气雷达数据能揭示探测天气目标的精细结构,对灾害性天气分析和预报预警至关重要。提高天气雷达反射率数据分辨率可以提升现有业务天气雷达对中小尺度强对流灾害性天气的监测和预警能力。本文在不改变雷达硬件的情况下,提出了基于注意力反向投影网络(Attention Back-Projection Network,ABPN)的天气雷达回波超分辨率重建算法用于提高雷达反射率数据分辨率。ABPN通过在深度反向投影网络(Deep Back-Projection Network,DBPN)中加入长短跳跃连接和通道注意力机制,对关键区域精细化重建结构特征。通过对实际天气过程超分辨率重建测试,结果表明,ABPN算法在雷达回波重建质量和主观视觉评估上有明显的优势,特别是在回波细节和天气雷达的边缘结构特征方面。

关 键 词:天气雷达  神经网络  超分辨率  反向投影网络  注意力机制
收稿时间:2022/9/6 0:00:00
修稿时间:2023/2/20 0:00:00

Algorithm for Weather Radar Echo Super-Resolution Reconstruction Based on Attention Back-Projection Network
YU Qiu,ZENG Qiangyu,ZHANG Fugui,WANG Hao,SHI Zhao,LI Haoran.Algorithm for Weather Radar Echo Super-Resolution Reconstruction Based on Attention Back-Projection Network[J].Meteorological Science and Technology,2023,51(3):319-330.
Authors:YU Qiu  ZENG Qiangyu  ZHANG Fugui  WANG Hao  SHI Zhao  LI Haoran
Abstract:High resolution weather radar data can reveal the fine structure of detected weather targets and are essential for catastrophic weather analysis, forecasting and warning. Improving weather radar reflectivity data resolution can enhance the monitoring and warning capability of existing operational weather radar for small and medium scale strong convective disastrous weather. In this paper, based on an attention back projection network (ABPN), the super resolution reconstruction algorithm is proposed to improve the resolution of weather radar reflectivity data without radar hardware modification. The attentional back projection network is accomplished by adding long and short skip connections in the deep back projection network (DBPN) and channel attention mechanism to refine and reconstruct structural features in critical regions. By testing the super resolution reconstruction on real weather processes, it is demonstrated that the ABPN algorithm has significant advantages in radar echo reconstruction quality and subjective visual evaluation, especially in terms of echo details and edge structure features of weather radar.
Keywords:weather radar  the neural network  super resolution  back projection network  the attention mechanism
本文献已被 维普 等数据库收录!
点击此处可从《气象科技》浏览原始摘要信息
点击此处可从《气象科技》下载免费的PDF全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号