STNLFFM

MODIS和Landsat表面反射率产品在空间和时间分辨率方面具有互补的特征。为了充分利用这些数据集，Gao等人 (2006) 开发了空间和时间自适应反射融合模型 (STARFM)。STARFM方法融合了来自MODIS的高频时间信息和来自Landsat的高分辨率空间信息，以30m的空间分辨率和每日时间分辨率生成合成的表面反射率产品。STARFM使用在相同日期收集的一对或多对Landsat-MODIS图像来预测其他MODIS观测日期在Landsat分辨率下的表面反射率。从那时起，已经开发了许多图像融合模型。Cheng等人 (2017) 提出了一种基于时空非局部滤波的融合模型 (STNLFFM)，该模型借助多时相粗分辨率反射率图像导出了多时相细分辨率反射率图像之间的新变换关系，并充分利用图像序列中的高度时空冗余来产生最终预测。但是，STNLFFM的计算性能一直是批量生产的瓶颈。
为了克服计算障碍并支持大尺寸图像的批量生产，我们设计并实现了基于计算统一设备体系结构 (CUDA) 的支持GPU的STNLFFM程序，称为cuSTNLFFM。通过利用GPU的大量并发计算线程，cuSTNLFFM可以大大减少计算时间并提高计算性能。实验表明，与在英特尔至强E3-1226 CPU上运行的顺序STNLFFM程序相比，使用Nvidia Tesla K40 GPU实现了100的加速。

MODIS and Landsat surface reflectance products have complementary characteristics in terms of spatial and temporal resolution. To fully utilize these datasets, Gao et al. (2006) developed the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM). STARFM fuses the high-frequency temporal information from MODIS and the high-resolution spatial information from Landsat to generate synthetic surface reflectance products with 30-meter spatial resolution and daily temporal resolution. STARFM uses one or more pairs of Landsat-MODIS images collected on the same date to predict surface reflectance at Landsat resolution for other MODIS observation dates. Since then, numerous image fusion models have been developed. Cheng et al. (2017) proposed a fusion model based on spatio-temporal non-local filtering (STNLFFM), which derives a new transformation relationship between multi-temporal fine-resolution reflectance images with the aid of multi-temporal coarse-resolution reflectance images, and makes full use of the strong spatio-temporal redundancy in image sequences to produce final predictions. However, the computational performance of STNLFFM has long been a bottleneck for batch production. To overcome the computational barriers and support batch production of large-size images, we designed and implemented a GPU-enabled STNLFFM program based on Compute Unified Device Architecture (CUDA), termed cuSTNLFFM. By leveraging the massive concurrent computing threads of GPUs, cuSTNLFFM can greatly reduce computational time and improve computational performance. Experiments show that using an Nvidia Tesla K40 GPU achieves a speedup of 100-fold compared to the sequential STNLFFM program running on an Intel Xeon E3-1226 CPU.

搜集汇总

背景与挑战

背景概述

STNLFFM数据集涉及一个基于时空非局部滤波的融合模型（STNLFFM），用于结合MODIS和Landsat卫星图像，以生成高时空分辨率的表面反射率产品。该数据集的核心是cuSTNLFFM程序，这是一个GPU加速版本，通过CUDA技术大幅提升计算效率，实验表明比CPU版本加速100倍。数据集由山西农业大学和中国气象局于2022年发布，提供相关代码和论文资源，旨在支持遥感图像融合的大规模应用。