A comparative assessment of multi-view fusion learning for crop classification
Published in IEEE International Geoscience and Remote Sensing Symposium (IGARSS), 2023
DOI: 10.1109/IGARSS52108.2023.10282138
Code: github.com/fmenat/MultiviewCropClassification
This work presents a comparative assessment of multi-view fusion strategies for crop classification using heterogeneous remote sensing data. The study addresses the challenges of combining sources with different resolutions, magnitudes, and noise characteristics, comparing traditional input-level fusion with more advanced fusion approaches. Experiments on the CropHarvest dataset show that multi-view fusion generally outperforms models based on individual data sources and previous fusion methods. However, no single fusion strategy consistently achieves the best performance across all scenarios. Instead, the optimal approach depends on the geographic test region. Based on these findings, the study proposes a preliminary criterion to guide the selection of suitable fusion strategies for crop classification.
Recommended citation: Mena, Francisco, et al. "A comparative assessment of multi-view fusion learning for crop classification." IGARSS 2023-2023 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 2023.
Download Paper | Download Slides