
The AIDurian team completed a structured multimodal data collection with partner farm **Belviz Farms**, capturing 189 Puyat durian fruits across multiple sensing modalities:
- 1,134 RGB images
- 1,134 thermal images
- 189 multispectral montages (8 NIR bands, 720–980 nm)
- ~378 acoustic recordings
Why this matters
Combined with the primary field dataset of 1,216 fruits collected across four harvest rounds (August 2024 – April 2025), this dual-dataset approach gives the AI models diverse training data spanning sensor modalities, imaging conditions, lighting environments, fruit origins, and ripeness stages — strengthening model generalization for real-world assessment.
Open access
The cleaned multimodal dataset is now publicly available on Zenodo under a CC BY 4.0 license: doi.org/10.5281/zenodo.18603796 — 23.4 GB across RGB, multispectral, thermal, and acoustic files, plus the fruit-characteristics CSV and full README.
