We used Sentinel-2 & point-based NFI data as sparse labels for weakly supervised CNN-based national LC mapping. We also showed that pretraining a self-supervised MAE before finetuning with NFI data improved model performance.
doi.org/10.3390/rs17...
We used Sentinel-2 & point-based NFI data as sparse labels for weakly supervised CNN-based national LC mapping. We also showed that pretraining a self-supervised MAE before finetuning with NFI data improved model performance.
doi.org/10.3390/rs17...
medium.com/@moraesd90/c...
medium.com/@moraesd90/c...