European Commission logo
English English
CORDIS - EU research results
CORDIS

An Artificial Intelligence Capable of Detecting Counterfeit Products Using a Cell Phone Camera

Periodic Reporting for period 2 - Microguard (An Artificial Intelligence Capable of Detecting Counterfeit Products Using a Cell Phone Camera)

Reporting period: 2021-07-01 to 2023-03-31

PROBLEM
Counterfeiting is a global scourge and a € trillion industry, responsible not only for economic losses such as revenue and jobs, but also threatening lives through counterfeiting of products such as electronics and pharmaceutical products. Each year, the EU loses €83 billion in sales (7.4% of all sales) due to counterfeiting and piracy. An additional €35 billion is also lost across the EU economy due to the indirect effects of counterfeiting and piracy, as manufacturers buy fewer goods and services from suppliers, causing knock-on effects in other areas. The counterfeiting industry is responsible for 2.6 million jobs lost globally during 2005 to 2013 and the figure could be as high as 5.4 million jobs lost by 2022. The EU loses 800,000 jobs every year. Moreover, counterfeit products can result in damaging the health of the customers. Counterfeit items do not go through the rigorous safety and compliance testing performed for legitimate products. This can lead to serious health issues and even death. For instance, the WHO estimated that 169,000 children died of pneumonia in 2018 due to ineffective counterfeit antibiotics. Counterfeiting operations are becoming increasingly sophisticated both in terms of the items produced and the complexity of their operational structure. Not only is it extremely difficult to adequately protect goods against counterfeiting, but the detection and mapping of the vast counterfeit operations that infiltrate global supply chains is near impossible due to lack of data. This prevents law enforcement agencies from fighting back against the counterfeiters to break up these operations.

SOCIETY IMPACT
Microguard and Microguard DeepTrace will protect revenues, jobs supported by these revenues, and save lives through the eradication of counterfeit goods. The EU is becoming increasingly vigilant towards the threat of counterfeiting and the misuse of security certification labels. It is estimated that 700,000 people die every year due to taking counterfeit medicine. The majority of these deaths are caused by the 20 most vulnerable medicines to counterfeiting. An example of Microguard’s contemporary impact is provided through a recent deployment of 115,000 labels to protect Incepta Pharma’s leading medicine Osicent 80, the biggest generic medicine in the world for lung adenocarcinoma, the most common type of lung cancer. A detailed calculation performed in collaboration with the customer shows that this limited deployment will save the lives of 1,625 people. Another exmple is the COVID-19 crisis, where over 31 million counterfeit medical face masks have fled the Chinese market in only 4 months. Microguard clearly addresses a current EU challenge and is being introduced at the optimum time to maximise both economic and social impact of the innovation.

OVERALL OBJECTIVES
- AI System Optimisation to reach an accuracy of 99% for 20 types of Lacoste care labels (currently 95%). Optimisation of AI system for mapping counterfeit operations; and development of user-interface and cloud analytics platform.
- Scale-up of Cloud Platform by re-writing existing cloud architecture to move from virtual server cloud to docker-based cloud; engineering of DeepTrace cloud architecture; load balancing and security tests to ensure compatibility with a large global userbase; and implementation of data architecture.
- Large scale demonstration of Microguard DeepTrace with companies from the existing pipeline, including Thales. Demonstration of newly optimised label-free authentication of customer packaging with exceptional accuracy on a large scale. Demonstration of AI-based mapping of a counterfeit operation in collaboration with World Customs Organisation. Review and implementation of trial outcomes and identified upgrades.
Task 1.1 Data collection and preparation (Status: In progress)
The data collection task was delayed due to delay in securing a data partner. Finally, the research POC agreement was signed with LACOSTE, giving access to all of their fakes.

Task 1.2 Optimization of the AI algorithm for direct label-free authentication (Status: Completed)
Microguard DeepTrace algorithm was optimized to recognise real and fake products. With the LACOSTE dataset, 98% accuracy has been achieved in detecting counterfeit LACOSTE polos with a smartphone camera.

Task 2.1: Engineering of cloud architecture (Status: Completed)
Rethinking of cloud architecture was completed. Switched all our servers to a docker based architecture. No parts of our cloud is on virtual servers and all is on dockers now.

Task 2.2: Software platform testing (Status: Completed)
The software platform deployed on the cloud was tested in preparation for the planned pilot demonstrations with potential customers. We also implemented an automatic testing of the server every few minutes with an automated phone call to our IT team when the servers aren’t responding using pager duty and dead man snitch.

Task 2.3 – Design and implementation of data architecture (Status: Completed)
As for our data analytics architecture, we switched to tableau, this allows for greater data analytics flexibility.
And to test all of these developments, a load test was performed by sending a lot of requests simultaneously to see how long the cloud would take to scale up and down.

Task 3.1 Detailed planning (Status: Completed)
A research POC was signed with LACOSTE. The detailed planning for the pilots including expectations, implementation, success criteria, project management and reporting was completed.
Label free authentication of counterfeit goods: There have never been any successful implementation of an artificial intelligence to detect counterfeit goods using smartphone pictures. Several universities or research entities have made attempts in this direction but none have been successful. The main issue being that none of these attempts managed to gather enough data to have a significant sample of the fakes of a given product. Cypheme has been successful in securing a research contract with LACOSTE, giving access to all of their fakes. The process is patented and 98% accuracy has been achieved in detecting counterfeit LACOSTE polos with a smartphone camera on known fakes while 50% accuracy on unknown fakes just by taking a single picture of the crocodile logo on the polo. Though the accuracy is not good enough for commercial viability but will be improved significantly before the end of the project.

The potential societal impact of this project are immense as counterfeiting is now 3.3% of world trade and kills more than malaria. If counterfeiters were a company, they would dwarf Amazon, Apple and Google. If a reliable AI can be deployed to detect counterfeit critical medicines, this could save hundreds of thousands of lives. Our current technology already saved hundreds of lives by detecting counterfeit anti cancer medicine, but the potential is much much bigger.
Lacoste logo imaging
auto testing
Server switching to docker based architecture
Data analytics architecture