Satellite Imagery Compressed to 64 Dimensions to Identify Tomato Fields
anselm · x · 2026-07-16
Google DeepMind's AlphaEarth Foundations can compress a year's worth of '10-meter per pixel' satellite imagery into 64 numerical values. Using this representation, a UC Davis team successfully identified all tomato fields in California.
The original text emphasizes that such tasks traditionally rely heavily on manual selection of color bands and seasonal features, requiring re-adaptation for different years or regions. The goal of the new method is to eliminate these fragile feature engineering steps, making surface crop mapping much more stable.
More from Research
- RSNA 2026 Deep Learning Labs return: eight hands-on AI classes over four days in Chicago — PRouzrokh · 2026-10-27
- Steering two orthogonal concept neurons makes a model compose "gnome" — Sauers_ · 2026-10-27
- Paper: data schedules and learning rate schedules should be designed together — alifcoder · 2026-10-27
- Health benchmark study: quality and diversity of eval items beat quantity — keunwoochoi · 2026-10-27
- Across 7 tasks, highest reasoning setting isn't the best for most open models — keunwoochoi · 2026-10-27
- Encrypted GPT-2 speedrun challenges FHE inference: best is 3.031 s/token, goal sub-1s — tensorqt · 2026-10-27