Spatio-temporal fusion is a feasible way to provide synthetic satellite images with high spatial and high temporal resolution simultaneously. Due to its practicability, spatio-temporal fusion has gotten increasing attention, for… Click to show full abstract
Spatio-temporal fusion is a feasible way to provide synthetic satellite images with high spatial and high temporal resolution simultaneously. Due to its practicability, spatio-temporal fusion has gotten increasing attention, for which many spatio-temporal fusion approaches have been developed. Most spatio-temporal fusion methods follow the “base fine image guided” (BFIG) fusion mode, resulting in the fact that their fusion results are similar to the base fine images. Therefore, these methods can perform well in the areas with limited surface changes due to the high similarity between the base and the predicted fine images. However, they might not be applicable in the areas with intense surface changes. In this paper we develop a Pansharpening-Based Spatio-Temporal Fusion Model (PSTFM) by introducing the Pansharpening fusion mode, which is “coarse image guided” (CIG), into spatio-temporal fusion. The PSTFM first trains a Pansharpening convolutional neural network (CNN), which then fuses the coarse images and reconstructed Pan images of the predicted time to recover the missing fine images. The newly proposed PSTFM is compared to three representative BFIG spatio-temporal fusion methods on two Landsat-MODIS datasets which both contain intense surface changes. After that, the experimental results are analyzed and discussed in detail. The experiments and the analysis demonstrate that the newly proposed PSTFM has remarkably qualitative and quantitative performance in predicting the intense surface changes while it is mediocre in the areas with low surface change intensity.
               
Click one of the above tabs to view related content.