Can a model trained on satellite data really find brambles on the ground?
17 hours ago
- #remote sensing
- #habitat mapping
- #machine learning
- Gabriel Mahler researched hedgehog habitat mapping using Agent Based Models (ABMs) and remote sensing.
- A bramble map was created by combining TESSERA earth representation embeddings with iNaturalist data.
- The model used an ensemble of logistic regression and a knn classifier.
- Field testing in Cambridge confirmed the model's accuracy in predicting bramble locations.
- Brambles were found in high-confidence areas, especially those not under cover.
- The model struggled with brambles partially obscured from above.
- Validation data collection included photos and GPS locations.
- A mobile phone-based active learning setup was suggested for future improvements.