Deep Dive Talk at FAIR Day 2025

Excited to share insights from our research on illumination-invariant hyperspectral imaging at the FAIR Day 2025!

The talk dives into the challenges of deploying AI in real vineyards and shows how our LISA framework keeps hyperspectral predictions stable under widely varying lighting conditions.

Highlights from the session:

• Why hyperspectral sensing is essential for precision viticulture and how RGB perception falls short
• Our Treebot platform: FX10 hyperspectral camera, UR5e arm, RTK-GPS, and custom vineyard datasets
• The Light-Invariant Spectral Autoencoder (LISA) architecture and its adversarial training strategy
• A full pipeline that pairs LISA with YOLO detection for real-time bunch tracking, yield, and quality prediction
• How illumination-robust models unlock scalable, automated decision-making for smart farming