Fusion of Sentinel-1 and Sentinel-2 Images for Classification of Agricultural Areas Using a Novel Classification Approach

摘要

A continuously growing world population increases steadily the demand of foods. This results in strong changes that occur on agricultural sites. Remote sensing data provides an excellent opportunity to monitor these changes which is a crucial base to asses the impact of these changes on the climate or the natural resources. In the presented study, we tested the performance of a new crop classification method for a stack of Sentinel 1 (S1) and Sentinel 2 (S2) images taken within one growing season. We proved, that the new PSP method performs better for S1 images revealing an overall accuracy (OA) of 75% compared to 60% for the Random Forest classifier (RF). The PSP method outperformed also for the fused dataset of S1 and S2 images (72% OA for PSP, 62% for RF). The results illustrate the benefits for crop classifications provided by PSP and give crucial insights for the advantages and limits of S1 and S2 data fusion.