Speaker
Description
Hyperspectral satellite missions create opportunities for plant health monitoring from space. The German Environmental Mapping and Analysis Program (EnMAP) is a spaceborne hyperspectral mission that provides dense spectral coverage for each ground pixel. This study proposes to use EnMAP data for plant pest and disease detection, an application area that remains largely unexplored. Previous remote sensing work has focused mainly on land-cover mapping, change detection, and assessment of crop or forest canopy condition. Higher spectral resolution has improved these tasks, but plant pathogen detection from satellites is still challenging. At the same time, advances in computer vision, particularly deep learning, now allow more powerful classification of complex hyperspectral signals. We propose a comparative study of three deep learning approaches for satellite-based crop and pathogen detection: convolutional neural networks (CNNs), Vision Transformers (ViTs), and hybrid CNN–ViT models. The planned study area is in the northeastern region of Botswana and covers approximately 450 km² of agricultural land. EnMAP Level-2A data (surface reflectance) will be used. The EnMAP-Box plugin in QGIS will handle data import and conversion from HDF5 to GeoTIFF. The same plugin will be used to set up classification experiments and manage training and validation workflows. Model performance will be evaluated using standard metrics. Overall accuracy, the Kappa coefficient, and the F1-score will provide complementary views of classification quality. These metrics will be used to compare CNN, ViT, and CNN–ViT models.
| Stream | Science or Engineering |
|---|