Researchers Use AI to Identify EVOO Provenance

A new way to deter­mine the authen­tic­ity of extra vir­gin olive oil has been devised by a group of researchers in Italy.

Their study, pub­lished by Food Chemistry, details a method that includes train­ing arti­fi­cial intel­li­gence to iden­tify the prove­nance of an extra vir­gin olive oil using its phe­no­lic com­pounds and sterols.

Researchers used Taggiasca Ligure extra vir­gin olive oil from Liguria in north­west­ern Italy.

“Still, the method­ol­ogy we deployed could apply to any other extra vir­gin olive oil, to any cul­ti­var, in any region,” Luigi Lucini, a researcher at the depart­ment for sus­tain­able food processes at the Università Cattolica del Sacro Cuore and co-author of the study, told Olive Oil Times.

The main dri­ver for the devel­op­ment was the spread of the Taggiasca olive, native to the region, to other coun­tries. Therefore, the researchers felt it was impor­tant to be able to iden­tify and label Taggiasca mono­va­ri­etals from Liguria.

“I hear that the Taggiasca cul­ti­var is being planted abroad, in places such as Greece,” Lucini said. ​“When we talk about wine, we are used to the con­cept of ter­roir. However, the link between extra vir­gin olive oil and the ter­ri­tory of ori­gin is a real thing, and it implies spe­cific qual­ity char­ac­ter­is­tics.”

In the study, researchers said they could cor­rectly iden­tify locally-pro­duced Taggiasca Ligure olive oils 100 per­cent of the time.

“We worked for four years on the project, and the last year was entirely spent train­ing the sys­tem and ver­i­fy­ing the effi­ciency of the method,” Lucini said.

The research team com­pared the new method to the FaceID authen­tic­ity tool widely adopted by smart­phone pro­duc­ers.

“That sys­tem learns to rec­og­nize dif­fer­ent angles of a spe­cific face to autho­rize access to the device,” Lucini said. ​“Our method does the same; instead of somatic para­me­ters, it rec­og­nizes chem­i­cal para­me­ters, allow­ing it to authen­ti­cate the pro­duc­t’s ori­gin.”

The researchers began the project by build­ing a robust dataset using 408 sam­ples of Taggiasca Ligure extra vir­gin olive oil col­lected from three har­vest sea­sons. With the coop­er­a­tion of local pro­ducer asso­ci­a­tions, they labeled every sam­ple with coor­di­nates.

Using metabolomics, the chem­i­cal fin­ger­print of a spe­cific cel­lu­lar process, researchers were also able to iden­tify thou­sands of dif­fer­ent com­pounds, dozens of which are unique to locally-pro­duced Taggiasca Ligure olive oil.

“Cholesterol-deriv­a­tives and phe­no­lics (tyrosols, oleu­ropeins, stil­benes, lig­nans, phe­no­lic acids and flavonoids) were the best mark­ers, based on sta­tis­tics,” the researchers wrote. ​“Our results strengthen the con­cept of ​‘ter­roir’ for extra vir­gin olive oil and indi­cate that pro­fil­ing sterols and phe­no­lics can sup­port extra vir­gin olive oil integrity if ade­quate data treat­ments are adopted.”

“Extra vir­gin olive oil con­tents vary from sea­son to sea­son,” Lucini added. ​“Especially in Liguria, where you can find olive trees grow­ing at sea level and oth­ers thriv­ing at hun­dreds of meters only a few kilo­me­ters away.”

“Differences might also come from weather or farm­ing tech­niques,” he said. ​“That is why we gath­ered data in dif­fer­ent sea­sons to deter­mine the exact mark­ers we needed.”

Once the data set was built, arti­fi­cial neural net­works were trained to iden­tify Taggiasca Ligure extra vir­gin olive oils and deployed to deter­mine the authen­tic­ity of oils labeled as such.

The researchers said the dataset should be flex­i­ble enough to iden­tify whether locally-pro­duced blends, includ­ing those claim­ing to com­prise the Taggiasca olive, really do.

“Just like the FaceID tool, which rec­og­nizes me even if I have glasses on, our method does the same, and it is not obso­lete when faced with olive oil blends,” Lucini said. ​“With FaceD, this hap­pens because wear­ing glasses is not a deter­min­ing para­me­ter. The same is true of our method.”

To test the sys­tem, the researchers tested blended olive oil sam­ples, where non-Taggiasca oils rep­re­sented any­where from 5 to 60 per­cent of the blend. Frantoio was one of the dif­fer­ent olive oils used in the blends.

“The rea­son we chose the Frantoio cul­ti­var is for the close sim­i­lar­ity of its genet­ics with the Taggiasca cul­ti­var, which is derived from the same olive tree as Frantoio,” Lucini said. ​“That is due to monks adopt­ing the olive trees dur­ing the Middle Ages, with the cul­ti­vars evolv­ing from there.”

“Both cul­ti­vars have a com­mon ances­tor, and if you use com­mon genet­ics analy­ses, the two cul­ti­vars are vir­tu­ally indis­tin­guish­able,” he added.

However, the researchers con­cluded that their arti­fi­cial intel­li­gence is ready to iden­tify Taggiasca Ligure extra vir­gin olive oils out­side lab­o­ra­tory set­tings. Now, the research team’s next steps will focus on wine.

“The rea­son is that we are try­ing to work on high added value prod­ucts which can jus­tify such demand­ing work,” Lucini said.

For this rea­son, the new method is mainly reserved for prod­ucts with a Protected Geographical Indication (PGI) or Protected Designation of Origin (PDO) cer­ti­fi­ca­tion from the European Union.

Italy boasts 49 extra vir­gin olive oils with a PDO or PGI cer­ti­fi­ca­tion, with sev­eral more can­di­dates seek­ing with own geo­graph­i­cal indi­ca­tors in the com­ing years.

“Extra vir­gin olive oil is one of the foods more exposed to fraud,” said Marco Trevisan, the research coor­di­na­tor and food chem­istry pro­fes­sor at Università Cattolica del Sacro Cuore.

“And it is even more true for pro­tected prod­ucts, such as Taggiasca Ligure, on which con­sumers are will­ing to spend more money,” he con­cluded. ​“Our work… is a rel­e­vant step for pro­tect­ing PDOs.”