基于TensorFlow实现FY-3D MERSI数据GPU并行快速几何校正  

GPU Parallel Fast Geometry Correction of FY-3D MERSI data based on TensorFlow

作　　者：王卫东 权文婷 王钊 周辉 WANG Weidong;Qüan Wenting;Wang Zhao;Zhou Hui(Shaanxi Meteorological Service Center of Agricultural Remote Sensing and Economic Crop,Xi’an 710016,China;Xi’an Jiaotong University,Xi’an 710049,China)

机构地区：[1]陕西省农业遥感与经济作物气象服务中心,陕西西安710016 [2]西安交通大学,陕西西安710049

出　　处：《遥感技术与应用》2023年第3期671-679,共9页Remote Sensing Technology and Application

基　　金：风云三号03批气象卫星工程地面应用系统区域特色遥感应用系统项目(FY-3(03)-AS-12.08);陕西气象高质量发展体系建设项目(141027020000210002)。

摘　　要：在TensorFlow框架下,有对多维数组进行GPU或CPU并行加速运算的特点,选择使用合适的经纬度投影模式,结合FY-3D MERSI L1数据自身带有高精度同分辨率的定位数据,按分辨率生成将经纬度数据对齐的张量(多维数组),计算生成新的像元映射位置信息,直接逐点映射对MERSI数据进行几何校正,同时可消除中分辨率极轨卫星因扫描观测和地球曲率带来的BowTie效应,再使用卷积运算计算反距离加权插值点数值,对几何校正后无数据像素点进行填充。使用这种方法,在TensorFlow框架下用Python语言实现了整个几何校正过程,并与ENVI软件的几何校正结果,对误差和校正精确度进行了对比测试,并且对几何校正的整体处理速度也进行了测试。结果表明本文算法与ENVI软件几何校正一致性高,5%绝对误差百分比以下的精确率大于0.92,结构相似SSIM指数在0.95左右,全部通道使用GPU并行加速完成几何校正的加速比大于36倍。这种方法既保证了几何校正的精度,而且整个数据处理过程快速高效。Based on the TensorFlow framework,which has the characteristics of GPU or CPU parallel accelera⁃tion calculation on tensor(multidimensional arrays).The WGS84 latitude and longitude projection is selected,combined with FY-3D MERSI L1 data with high-precision and same-resolution positioning data,to generated tensors(multidimensional array)to align latitude and longitude data according to resolution,and calculated the new image pixel mapping position information.According to the position index information,the MERSI data can be geometric corrected point by point,and the BowTie effect caused by the scanning observation and earth curvature of medium-resolution polar orbit satellites can be eliminated at the same time.Finally,the convolu⁃tion is used to calculate the inverse distance weighted interpolation point values and fill the pixels with no data af⁃ter geometric correction.Using this method,the author implemented geometric correction of all 25 channels of FY-3D MERSI data in Python under the TensorFlow framework.Compared with the geometric correction re⁃sults with ENVI software as the standard.The error and correction precision are calculated,and the overall pro⁃cessing speed of geometric correction is also tested.The results show that the algorithm proposed in this paper has a high consistency with the geometric correction of ENVI software,and the accuracy below 5%absolute er⁃ror percentage is greater than 0.92,and the structural similarity SSIM index is around 0.95.Speedup is more than 36 times to complete geometric correction for all channels using GPU parallel acceleration.In summary,the geometric correction method adopted in this paper is fast and efficient in processing,and ensures the accura⁃cy of correction.

关 键 词：FY-3D MERSI 几何校正 反距离加权插值 TensorFlow 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]