Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-2008
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-10-20T10:42:54.320424Z
status: published
lens: g_space
sector: business
headline: Deep chemometrics for retrospective auditing of the water-nutrient nexus in olive oil via 1D-CNN and explainable AI
dek: Transitioning toward high-value functional food markets increases demand for forensic tools capable of verifying the "water-nutrient nexus", the complex agronomic interplay between irrigation and fertilization that shapes olive oil composition and authenticity. Conventional chemometrics often fail to capture synergistic, non-linear cultivation fingerprints within complex chemical matrices. This study introduces a "Deep Chemometrics" framework for retrospective auditing of olive oil history using FTIR spectroscop…
gain_title: A 1D-CNN trained on FTIR spectra classified nine irrigation and fertilization regimes of olive oil, giving a transparent way to audit cultivation history and support supply-chain authenticity.
problem_title: (none)
trace_subject: (none)
gain_reading: A 1D-CNN trained on FTIR spectra classified nine irrigation and fertilization regimes of olive oil, giving a transparent way to audit cultivation history and support supply-chain authenticity.
gain_evidence: a transparent, scientifically interpretable, and robust strategy for retrospective authentication of olive oils | supporting supply-chain authenticity
problem_reading: (none)
problem_evidence: (none)
quick_read: Researchers built a deep chemometrics framework that reads infrared spectra of olive oil to infer how the olives were irrigated and fertilized. The network was trained on 324 spectra covering nine treatment combinations, and physics-informed augmentation was used to offset the small sample set.

The approach matters because high-value food markets need tools to verify claims about how products were grown. Explainable AI tied the model's decisions to specific chemical bands, which makes the results easier to trust. The held-out accuracy is lower than the validation accuracy, and the study does not report results beyond its sampled regimes.
limitation: Performance dropped from 97.89% validation accuracy to 83% on the held-out test set, and the study needed data augmentation because sample availability was limited.
tag: Evidence-backed gain
key_points: The dataset comprised 324 FTIR spectra from nine irrigation by fertilization regimes. | Validation accuracy reached 97.89%, while held-out test accuracy was 83%. | The test result was a 25.8% improvement over the baseline model. | Explainable AI highlighted the 1744 cm-1 and 2922 cm-1 bands as the main drivers of classification. | Kruskal-Wallis tests found significant treatment differences in those bands (p = 0.0076 and p = 0.0025).
rundown: The study frames the problem as the difficulty of verifying the water-nutrient nexus, the interplay between irrigation and fertilization that shapes olive oil composition. Conventional chemometrics often fails to capture the non-linear, synergistic fingerprints in complex chemical matrices.

The customized 1D-CNN had macro-averaged precision of 0.85, recall of 0.83 and an F1-score of 0.83 on the test set. Gradient-based explainability pointed to the ester carbonyl stretch and the CH2 asymmetric stretch as the key spectral features.

Peak deconvolution with constrained pseudo-Voigt fitting and non-parametric statistics were used as independent spectroscopic validation of the CNN's feature importance.
sources:
- peer_reviewed | Food Chemistry | https://doi.org/10.1016/j.foodchem.2026.151546 | 2026-10-13
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

dedf131d0a7d3ab1c9999811403151cb518aff547be069c331fad1e67d730fb1

previous

0000000000000000000000000000000000000000000000000000000000000000