The Rosselkhoznadzor has developed an artificial intelligence model that can detect violations in all production chains of animal products in order to reduce the proportion of counterfeit and falsified products of animal origin on the market as much as possible. The innovation is beneficial to a healthy society, conscientious agricultural producers and the state.
Artificial intelligence methods for detecting violations have not been previously applied by the Rosselkhoznadzor.
Innovation Summary
Innovation Overview
The introduction of systems which help surveillance bodies to identify all stages of production and movement of food products - from farm to fork - is an international trend. It allows to control the circulation of food products at the state level, thereby protecting the final customer from purchasing low-quality and potentially dangerous goods.
The Federal State Information System “Mercury” (FSIS “Mercury”) was created by the Rosselkhoznadzor to ensure the traceability system of all animal products in the Russian Federation, both domestically produced and imported. The work of this system is regulated by the Federal Law “On Amendments to the Law of the Russian Federation “On Veterinary Medicine” and certain legislative acts of the Russian Federation” dated July 13, 2015 No. 243-FZ. The movement and circulation of products without registering of all actions in the FSIS “Mercury” is prohibited in accordance with the specified regulations.
FSIS “Mercury” accepts an electronic veterinary certificate or an electronic veterinary accompanying document (eVAD), which is issued at each stage of controlled products movement, as the basis of traceability. The list of FSIS “Mercury” controlled products includes meat and meat products, fish and seafood, food products, honey, non-food products of animal origin, feed and feed additives, live animals. The following categories of data can be traced by eVAD and their chains for products of all the above mentioned types: place of origin, date of production, manufacturer, supplier, volumes of incoming and outgoing products, ratio of types and quantities of raw materials to finished products, expiration dates, transportation and production sites.
As a result, traceability of each batch is created at all stages of the production chain.
As of today, more than 1 million participants in the agricultural product market are registered at the FSIS “Mercury”, creating more than 240 million eVADs each month.
Specialists from a monitoring group of 150 inspectors carry out analysis of electronic veterinary certificates and all the production chains of products of animal origin located in FSIS “Mercury”. Currently, the monitoring group can analyse only 0.03% of the total number of the products’ movement by eVAD, therefore, about 470 thousand inspectors will be required to analyse 100%!
An artificial intelligence model was developed to fully control the eVAD and their chains, its operation will ensure the control over 100% of eVAD, and the number of inspectors will decrease to 54 people (by three times).
Monitoring within the framework of eVAD full control using an artificial intelligence model will minimize the risks of counterfeit and falsified food products incidents in the territory of the Russian Federation, ensure the biological safety of raw materials, as well as products of animal and vegetable origin, increase the level of regulatory and technical support for the production and circulation of agricultural products, increase the export attractiveness of Russian food produce, reduce the dependence on using dubious technologies and raw materials in food production, and increase consumer demand and the quality of life.
The experience gained in implementation of artificial intelligence into the FSIS “Mercury” is supposed to be extended to the traceability system of plant products “Argus-Fito”, which will improve the quality and safety of plant products.
Innovation Description
What Makes Your Project Innovative?
Detection of counterfeit and falsified products through an analysis of electronic veterinary certificates by specialists of monitoring groups ensures processing of only 0.03% of the entire data array accumulated in the FSIS “Mercury” per month and, as a result, has a low percentage of violations detection in relation to their total number.
The use of artificial intelligence in the FSIS "Mercury" significantly increases the number of violations detected, which allows to increase the labour productivity by 8600 times and to reduce the cost of processing one eVAD from 300 Rubles to 3.5 Kopecks.
What is the current status of your innovation?
The model has been tested. Currently, the model code is being integrated in the FSIS “Mercury”.
Innovation Development
Collaborations & Partnerships
- Alekseeva Svetlana Aleksandrovna, Deputy Head of the Federal Service for Veterinary and Phytosanitary Surveillance - work on the project;
- Konstantin Arkadievich Savenkov, Deputy Head of the Federal Service for Veterinary and Phytosanitary Surveillance - work on the project;
- industry divisions of the Rosselkhoznadzor - results examination;
- information and analytical department of the Rosselkhoznadzor - technology development and methodology of the AI model.
Users, Stakeholders & Beneficiaries
Representatives of government authorities and civil society institutions can carry out control and supervisory activities more effectively, which entails an increase in food and biological security.
Companies can make big profits by improving the quality of raw materials and products (reducing reputation risks), eliminating unscrupulous market participants (reducing dumping).
Citizens can receive better products, raising the standard of living.
Innovation Reflections
Results, Outcomes & Impacts
The Rosselkhoznadzor has accumulated almost 40 TB of high quality data on the movement of raw materials and production.
Based on a sample for 3 months (600 million records), a relevant model was obtained which processed a 3-month data array in 2 hours. The AI not only confirmed the patterns of falsification of products identified by experts, but also revealed additional 15-20 suspicious patterns to each of the previously established ones. The accuracy of the AI model was 78 gini. With further development and training of AI, the accuracy can reach 90 gini out of 100 possible.
Challenges and Failures
The main problem is to ensure data cleanliness. Additional filters were used to bring the data into proper condition, which made it possible to train the AI model correctly.
Conditions for Success
The conditions for successful development of the innovation include:
- creation of infrastructure, which will be further integrated into various AI systems;
- development of technology in the field of AI.
Lessons Learned
The analysis of a large amount of information should be done using programmed models due to their greater productivity.
Experts' main task is to formulate selection criteria for data of interest, which engineers will turn into an algorithm and set into a model.
The use of artificial intelligence to detect anomalies is optional in the case of the possible presence of non-obvious signs of the required information.
Status:
- Implementation - making the innovation happen
Date Published:
15 January 2021