基于傅里叶神经算基于傅里叶神经算子的遥感数据预测方法子的遥感数据预测方法
- 分享到:
摘要：
地球遥感是气象卫星的主要任务。由于受云层遮挡、宇宙射线辐射等因素影响，气象卫星所获取的遥感数据通常存在大量缺失及异常。傅里叶神经算子具有效率高、精度高、分辨率灵活等特性，基于此，提出一种基于傅里叶神经算子的遥测数据预测算法。该算法首先对遥感数据缺失值利用空间均值法和拉格朗日插值法进行填充，之后用傅里叶神经算子训练出空间数值在时间域上的映射关系，最后利用训练出来的模型对于最新的遥感数据进行预测，基于风云4号遥感卫星真实遥感数据的仿真实验结果表明，所提出的方法在较长期的时序预测中仍能保持较好的预测精度。
Remote sensing of the earth is the main task of meteorological satellites. Due to factors such as cloud cover and cosmic ray radiation, remote sensing data obtained by meteorologiea satellites often have a large number of missing and abnormal data. Fourier neural operators have the characteristies of high efficieney, high accuracy, and flexible resolution. This paper proposes a remote sensing data prediction algorithm based on Fourier neural operators. The algorithm first fills in the missing values of remote sensing data using the spatial mean method and Lagrange interpolation method, and then trains the mapping relationship of spatial values in the time domain using Fourier neural operators. Finally, the trained model is used to prediet the latest remote sensing data., Simulation experiments based on real remote sensing data of Fengyun-4 remote sensing satelite show that the method proposed in this paper is more effective than others. Good prediction accuracy can still be mainlined in long-term time series forecasting.
作者：
卫兰，朱建璇，徐晓斌，范存群，林曼筠，赵现纲
Wei Lan,Zhu Jianxuan,Xu Xiaobin,Fan Cunqun,Lin Manyun,Zhao Xiangang
机构地区：
中国气象局中国遥感卫星辐射测量和定标重点开放实验室/国家卫星气象中心(国家空间天气监测预警中心)；许健民气象卫星创新中心；北京工业大学信息学部
引用本文：
卫兰，朱建璇，徐晓斌等。基于傅里叶神经算子的遥感数据预测方法[J].Betway官方客服学报(自然科学版),2025,53(1):82-91.(Wei Lan,Zhu Jianxuan,Xu Xiaobin,et al.Prediction method of remote sensing data based on fourier neural operator[J].Journal of Henan Normal University(Natural Science Edition),2025,53(1):82-91.DOI:10.16366/j.cnki.1000-2367.2023.06.30.0002.)
基金：
国家重点研发计划项目；北京市自然科学基金；风云卫星应用先行计划
关键词：
地球遥感；时序预测；长期预测；傅里叶神经算子
earth remote sensing;time seres prediction; long-term prediction;fourier neural operator
分类号：
TP79
