terclim by ICS banner
IVES 9 IVES Conference Series 9 OIV 9 OIV 2024 9 Short communications - Safety and health 9 Mineral-wine profile and AI: wine authentication and identification

Mineral-wine profile and AI: wine authentication and identification

Abstract

The use of a wine’s mineral profile (MWP) as a stable and distinctive fingerprint can revolutionize wine authentication. By employing inductively coupled mass spectrometry, we assessed the concentration of 11B, 23Na, 24Mg, 27Al, 28Si, 31P, 34S, 35Cl, 39K, 43Ca, 45Sc, 47Ti, 51V, 52Cr, 55Mn, 56Fe, 59Co, 60Ni, 63Cu, 66Zn, 75As, 79Br, 85Rb, 88Sr, 89Y, 90Zr, 93Nb, 111Cd, 118Sn, 127I, 133Cs, 137Ba, 139La, 140Ce, 141Pr, 146Nd, 182W, 205Tl, 208Pb, 238U with minimal sample preparation. MWP is shaped at least by soil composition and winemaking techniques. This last factor may overshadow the terroir when defining this profile, hindering origin-related information extraction. However, the integration of artificial intelligence (AI) presents itself as a solution. More than 19,000 MWPs were analysed, laying the groundwork for a machine-learning algorithm to assess wine’s country, region and main grape variety. Extreme Gradient Boosting was employed, exceeding scores of areas under the receiving operating characteristic curves of 0.9 for country, French wine region and grape variety classification. This performance enables a specificity of wine authentication up to 99%, demonstrating the potential of combining AI and MWP analysis. This study highlights the importance of comprehensive MWP datasets for advancing AI applications in origin verification, offering a promising tool for the wine industry to enhance security and consumer trust.

Profilo multiminerale e intelligenza artificiale: autenticazione e identificazione del vino

Migliorare il profilo multiminerale dei vini: dall’autenticazione all’identificazione con l’intelligenza artificiale per una maggiore sicurezza.  L’analisi del profilo di concentrazione minerale di un vino fornisce un’impronta digitale distintiva per ogni cuvée. A differenza dei profili organici, questa firma di identificazione (profilo caratteristico) rimane stabile nel tempo e può essere decifrata con un metodo di analisi diretta basato sulla spettrometria di massa ad accoppiamento induttivo (icp-ms). Con pochi millilitri di vino e una rapida preparazione, è possibile identificare più di 40 elementi minerali in concentrazioni che vanno dai grammi per litro per il potassio alle decine di nanogrammi per litro per le ultratracce metalliche, come i lantanidi.  Questa metodologia ci permette di caratterizzare quello che chiamiamo il wine mineral profile (mwp). Gli elementi minerali svolgono un ruolo cruciale nel territorio del vino, essendo principalmente derivati dal suolo attraverso l’uva e influenzati da varie tecniche di vinificazione. Tuttavia, nonostante la loro importanza, le caratteristiche del suolo sono spesso oscurate dalla moltitudine di procedure enologiche successive, ponendo notevoli sfide per l’estrazione di informazioni relative all’origine in un contesto convenzionale. Il nostro studio dimostra che l’intelligenza artificiale (ia) sta emergendo come uno strumento ottimale per decifrare accuratamente le informazioni sull’origine dal profilo minerale del vino, a condizione che venga misurato un numero sufficiente di elementi minerali e che venga esaminato un insieme ampio e completo di campioni di vino per un apprendimento efficace.  Nel corso di questo studio, in poco più di un anno è stato creato un set di dati comprendente più di 18.000 mwp. La nostra analisi si è poi concentrata sullo sviluppo di un metodo di apprendimento automatico per valutare l’origine del vino (paese, regione e denominazione) e il vitigno principale. Sono stati testati sei modelli confrontando l’area sotto la curva roc (auc). Sono stati raggiunti punteggi medi di auc superiori a 0,9 per la classificazione del paese, della regione vinicola francese e del vitigno per le identificazioni multiple di “vini sconosciuti”. Questo studio rappresenta la prima indagine completa su questa scala che coinvolge campioni di vino, sottolineando l’importanza di un set di dati mwp completo per le applicazioni di ia nella verifica dell’origine del vino. L’autenticazione del vino con una specificità superiore al 99% può già essere offerta utilizzando questo approccio.

Profil multi-minéral et ia : authentification et identification des vins 

Valorisation du profil multi-minéral des vins : de l’authentification à l’identification par intelligence artificielle pour une sécurisation renforcée.  L’analyse du profil de concentration en éléments minéraux d’un vin offre une empreinte distinctive pour chaque cuvée. Contrairement aux profils organiques, cette signature d’identification reste stable dans le temps et peut être déchiffrée grâce à une méthode d’analyse directe par spectrométrie de masse à couplage inductif (icp-ms). Avec quelques millilitres de vin et une préparation rapide, plus de 40 éléments minéraux peuvent être identifiés dans des concentrations allant de l’ordre du gramme par litre pour le potassium, à des concentrations de quelques dizaines de nanogrammes par litre pour les ultra-traces métalliques, telles que les lanthanides. Cette méthodologie permet de caractériser ce que nous appelons le profil minéral du vin (mwp). Les éléments minéraux jouent un rôle crucial dans le terroir du vin, étant principalement issus du sol à travers les raisins et influencés par diverses techniques de vinification. Cependant, malgré leur importance, les caractéristiques du sol sont souvent occultées par la multitude de procédures œnologiques ultérieures, posant ainsi des défis considérables pour extraire des informations liées à l’origine dans un contexte classique. Notre étude démontre que l’intelligence artificielle (ia) émerge comme un outil optimal pour décrypter avec précision les informations d’origine à partir du profil minéral du vin, à condition qu’un nombre suffisant d’éléments minéraux soient mesurés et qu’un ensemble de données volumineux et complet d’échantillons de vin soit examiné pour un apprentissage efficace. Au cours de cette étude, un ensemble de données comprenant plus de 18 000 mwp a été constitué en un peu plus d’un an. Notre analyse s’est ensuite concentrée sur le développement d’une méthode d’apprentissage automatique pour évaluer l’origine du vin (pays, région et appellation) et le cépage principal. Six modèles ont été testés en comparant l’aire sous la courbe roc (auc). Des scores moyens d’auc supérieurs à 0,9 pour la classification par pays, pour la région viticole française et pour le cépage ont déjà été atteints pour de multiples identifications de « vins inconnus ».  Cette étude représente la première investigation complète à cette échelle impliquant des échantillons de vin, soulignant l’importance d’un ensemble de données mwp complet pour les applications d’ia dans la vérification de l’origine du vin. L’authentification d’un vin avec plus de 99% de spécificité peut déjà être proposée grâce à cette approche.

DOI:

Publication date: November 18, 2024

Issue: OIV 2024

Type: Article

Authors

Leticia Sarlo1,2, Coraline Duroux2, Théodore Tillement2, François Lux1,3, Olivier Tillement1

1 Institut Lumière-Matière, UMR 5306, Université Claude Bernard Lyon 1-CNRS, Université de Lyon, Villeurbanne Cedex 69100, France
2 M&Wine, 305 rue des Fours, 69270 Fontaines Saint Martin, France
3 Institut Universitaire de France (IUF), Paris

Contact the author*

Tags

Full papers OIV 2024 | IVES Conference Series | OIV | OIV 2024

Citation

Related articles…

Les outils pour favoriser le renouvellement des générations en viticulture

French lawmakers have chosen the family-type winegrowing business as the benchmark for drafting the legal framework for winegrowing businesses and winegrowers. In france (source: msa), in 2022, there were 1,444 new winegrowers, an increase of 3% compared with 2021, representing 10% of new farm managers. The retention rate for winegrowers is 75% (up 13% on 2021), compared with 77% for all agricultural sectors (stable).

The evolution of italian vine nursery production over the past 30 years

Italy has a long history of viticulture and has become one of the world’s leading producers of vine propagation material. The Italian vine nursery industry is today highly qualified and has become highly competitive on a global scale. The quality of the material is guaranteed by compliance with European Union regulations, which have been in force since the second half of the 20th century and have subsequently been supplemented and updated.

“Gentle” sustainable extraction from whole berry by using resonance waves and slight over CO2 overpressure

The traditional methods of grape extraction of enochemical compounds use very often mechanical energy by pistons such as the pigeage or mechanical energy produced by must (delestage, pumping over). Recent trend by winemaker is trying to introduce in the fermentation tank, whole berry grape to avoid even minimal oxidation. Unfortunately, the use of the traditional mechanical techniques aforementioned, very often do not guarantee the optimal extraction with residual sugars in the marc. Use of resonance waves (airmixingtm) and a slight overpressure by CO2 (adcftm) permit to work on whole berry guaranteeing the perfect extraction.

La producción vitivinícola como fuente de impactos positivos en el medio seminatural

Wine is at risk of being labelled as a dangerous health product, based on studies that focus on its alcohol content. However, multiple studies suggest that moderate consumption is healthy. Changing the focus from health impacts to the environmental and socioeconomic impacts that wine companies cause, what can be said?

Hyperspectral imaging for precision viticulture

Precision viticulture aims to optimize vineyard management by monitoring and responding to variability within vine plots. this work presents a comprehensive study on the application of hyperspectral imaging (hsi) technology for monitoring purposes in precision viticulture. authors explore the deployment of hsi sensors on various platforms including laboratory settings, terrestrial vehicles, and unmanned aerial vehicles, facilitating the collection of high-resolution data across extensive vineyard areas.