DSensD+: Decision-level Sensor Dropout with Mutual Distillation for classification tasks

Public repository of our work Multi-sensor model for Earth observation robust to missing data via sensor dropout and mutual distillation

---

The DSensD+ method is shown in the previous image, while in our research work DSensD is also introduced. This is a simplified version using only Sensor Dropout at the decision-level. We focus on classification tasks in the Earth observation domain.

Note

Read about the used data in data folder

Training

• To train our novel DSensD+ (decision-level sensor dropout with mutual distillation) run

python train_multi.py -s config/dsensdp_ex.yaml

• To train our DSensD (decision-level sensor dropout) run

python train_multi.py -s config/dsensd_ex.yaml

• To train the baseline FSensD (feature-level sensor dropout) run

python train_multi.py -s config/fsensd_ex.yaml

• To train the baseline ISensD (input-level sensor dropout) run

python train_single.py -s config/isensd_ex.yaml

Note

Other competitors were used from their original code and also following our previous work at CoM-views.

Evaluation

• To evaluate the model by the prediction performance:

python evaluate.py -s config/eval_ex.yaml

Installation

Please install the required packages with the following command:

pip install -r requirements.txt

🖊️ Citation

Mena, Francisco, et al. "Multi-sensor Model for Earth Observation Robust to Missing Data via Sensor Dropout and Mutual Distillation." IEEE Access, 2025.

@article{mena2025dsensd,
  title = {Multi-sensor Model for Earth Observation Robust to Missing Data via Sensor Dropout and Mutual Distillation},
  author = {Mena, Francisco and Ienco, Dino and F. Dantas, Cassio and Interdonato, Roberto and Dengel, Andreas},
  journal = {IEEE Access},
  year = {2025},
  doi = {10.1109/ACCESS.2025.3568706},
  publisher={IEEE},
  volume={13},
  pages={83930 -- 83943},
}