Using AI to Spot Deforestation-related Cows on Satellite Images
summary
Claims ability to detect and count cattle from 30 cm RGB satellite tiles across the Brazilian Amazon, using PyTorch to train a two-step neural network that pairs a binary herd classifier with a density-estimation model. Training set expands to ~5,000 heterogeneous tiles; evaluation reports mean absolute errors around ~0.8 on earlier data and improved accuracy after retraining and ensembling. System produces heatmaps whose summed pixels estimate herd size and is already used to flag incursions into protected areas.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:31Motivation for Amazon deforestation research
- 5:45Animal counting in satellite imagery
- 8:18Detecting cattle on satellite images
- 9:44Density-based CNN model
- 11:12Training data and model development
- 13:41Model limitations and false positives
- 16:11Evaluating model performance
- 18:24Improving model with heterogeneous data
- 21:16Cattle density estimation applications
- 24:09Model improvements and future work
- 25:36Q&A
related talks
36mWatching Millions of Trees with Python: For Impact and Profit
36mClassification of Satellite Imagery
28mGeoAI: Evolution of AI-Based Tree/Vegetation Detection Models Using Imagery
27mSmart Forestry: AI in the Service of Nature
48mTaking the planet’s pulse: AI to empower adaptation to global change
32mMeasuring Vegetation Health to Predict Natural Hazards
21mUnderstanding World food economy with satellite
18mDeep learning for Space imagery applications