Supply Chain Integrity

To boost the value of Australia’s exports and build on our brand, we need to validate the quality, safety and provenance of our products. Using the red meat sector as a use case, CSIRO is developing technologies that can validate claims about the origin of a product, its authenticity and adherence to ethical production practices, as well as enhancing privacy and efficiency for producers​.

Trusting data collected on the farm

To validate provenance claims at the start of the supply chain, we need to collect data from sensors and devices on animals and on the farm. Data from these devices should not be trusted by default, but a reputation for trust in that data can be created through corroborating evidence. Our trust in temperature and location data collected from a smart device on an animal can be enhanced using weather data from BOM and location data from nearby animals. To ensure data trust, CSIRO is designing a system that will reliably collect, validate and securely communicate and store data from multiple devices on the farm and from multiple sources in the cloud.

Reducing regulatory burden through automated compliance

By automatically verifying compliance against standards and guidelines, we can replace manual checks and reduce regulatory burden for producers. Working with domain experts, CSIRO has transformed a set of industry rules on animal handling, water access, feeding and animal transport into quantifiable data that can be used to determine compliance automatically. CSIRO has developed systems that can identify and monitor animal behaviour. For example the data collected by an on-ear device can determine if that animal has access to enough water complying with industry welfare standards.

Identifying the biological origin of an agricultural product

CSIRO is investigating techniques to confirm the authenticity of a claim that a product comes from Australia, or a specific region in Australia. Using Australian beef and cherries, we are testing which unique biogeochemical signatures or markers (including geochemical, biochemical, isotopic, lipidomic and genomic) can be used to validate biological origin. This information will be used to develop a model that can predict the biological signature of a product from a distinct growing region. To maintain privacy and protect commercial interests, we are developing a model that ensures unique information about an individual farm cannot be deduced from the data.

Tracking individual cuts of meat

Traceability in the red meat supply chain goes beyond the farm. As individual cuts of meat are being handled in a processing facility it can be challenging to keep track of each piece and trace it back to the source animal. As individual cuts of meat are being handled in a processing facility it can be challenging to keep track of each piece and trace it back to the source animal. CSIRO has developed a computer vision algorithm that allows traceability of individual cuts of meat through aggregation and disaggregation during processing.

Detecting fraud in the supply chain

To detect irregular activity as goods move through the supply chain, CSIRO is developing new methods to analyse the rich data in digital supply chains. Sensors capture motion, temperature and location data to characterise a normal transport route, enabling detection of potentially fraudulent activities when they occur and distinguishing between different types of irregular events. We are investigating using green sensors that generate their own energy replacing the need for a battery, making it a low-cost option for industry.


Advancing Supply Chain Integrity Workshop held online on 2 December 2020

Advancing Supply Chain Integrity Introduction. Recording.
Dana Sanchez, Ryan McCallister, Aaron InghamBeef exports are valued at more than $10 billion per year, so it is important that Australia delivers a premium, safe and healthy product to the world to maintain this market share. Australian producers are renowned for very high standards of animal health and welfare, and environmental stewardship. However, in a fragmented and complex supply chain this message is easily lost.
Industry Panel Session. Recording.
Panel Members: Lucinda Corrigan, Sonja Dominik, Ian Jenson, Skye Richmond

Facilitator: Mark Hedley

In September, MLA commissioned a report analysing product integrity systems in the Australian red meat industry. Our panel members will discuss a key finding from the report –

‘The industry views on the need for investment in enhanced supply chain integrity systems are at odds with those of researchers and technology/service providers.’

Advances in the use of biological origin to verify authenticity. Recording.
Uta Stockmann, Sonja Dominik, David SmithTagless verification of provenance and identification of product is possible through unique signatures, biogeochemical, isotopic, lipidomic and genomic. In an Australian context, we show two levels of granularity – biogeochemical and isotopic signatures enabling verification to a region and even to the property of origin; and genomic signatures enabling unique identification of individual animals. In addition, we demonstrate privacy-preserving signature reporting for the accurate prediction of regional product provenance while providing confidentiality to primary producers.
Maintaining visibility of products moving through the supply chain and detecting fraud in an affordable way. Recording.
Sara Khalifa, Peter BaumgartnerWe report on our approach to tracking and anomaly detection in a boxed meat transport chain. We focus data collection from energy-harvesting sensors and analysing them with machine learning and rule-based methods.
Automated tracking of individual cuts of meat back to the source animal. Recording.
Dadong WangHow can we automatically track individual prime cuts of meat in the deboning room of an abattoir using AI based video analysis? We will present our methods and some initial results to address this problem.
Panel Session: Re-thinking blockchain – hype or need? Recording.
Panel members: Charles Morris-Turner, Volkan Dedeoglu, Sherry XuFacilitator: Phil Valencia
Demonstrating Automated Farm Provenance. Recording.
Phil Valencia

Export value built on Australia’s brand will need data to validate the safety, production and origin of our products. But how can we trust the data and ensure it can’t be modified? And how can we fulfil government requirements without being overwhelmed by regulatory burden? Provenance, Trust and Automated Compliance are proposed and demonstrated as technologies towards achieving this.

The technology components of the Automated Farm Provenance Demonstrator are listed in the rows below.

Reducing regulatory burden through automated compliance checking. Recording.
Nick van BeestWe present a regulatory technology for checking compliance with welfare guidelines and ensuring animal well-being. We show how to track animal behaviour and link that to the welfare guidelines and standards, so that potential issues are automatically detected and farmers are notified.
Low power cattle behaviour classification on an ear tag device. Recording.
Reza ArabloueiWe present the data processing pipeline and machine-learning algorithms developed for classifying cattle behaviour on the embedded system of the ear tags. The behaviour classification results are utilized by the trust and regtech components.
Video-based identification and classification of cattle behaviours. Recording.
Chuong NguyenWe present our data annotation and classification pipeline to recognise individual cows and their behaviour (drinking, grazing and other) from on-farm videos. Annotations and/or classifications from videos are used as ground truth to label other synchronised sensor streams (from the embedded system of the ear tags) to train machine learning algorithms relying only on sensor streams to recognise cattle behaviour.
High-resolution herd dynamics tracking for social and welfare metrics. Recording.
We report on recent on-farm trials, where our refined GPS devices are able to record and report in real-time the herd dynamics of multiple sheep flocks. Machine learning techniques are designed to identify the leaders of the herds and understand the stability of the herds and the social needs of individual animals.
Ensuring data and devices on the farm are protected and trusted. Recording.
Dongxi LiuWe introduce the EnerID blockchain, which is accessible for everyone (individuals or agencies) to record any big or small information without cost. EnerID can process messages in different efficiency modes and has been patented by CSIRO.  In addition, two lightweight encryption schemes are introduced, one of which can generate very short ciphertexts for short messages.
On-farm data trust and provenance. Recording.
Volkan DedeogluWe present our on-farm data trust and data provenance architectures. Our data provenance architecture is built on IoT edge blockchain and distributed data storage technologies, while data trust is established by our novel trust and reputation mechanisms.
Treating cows badly: A high fidelity simulator for Automated Farm Provenance. Recording.
Karl Von RichterWe present a Multi-Agent Systems Simulator (MASS) engine for simulating large scale distributed IoT networks of animals with smart ear tags on smart farms and pushing synthetic data through the demonstrator. Simulation of the animals and the smart devices they interact with allows us to test our technologies without physical subjecting animals to unethical conditions or needing to deploy devices on the farm.

Contact

Skye Richmond

Business Development & Commercialisation

skye.richmond ‘at’ data61.csiro.au

This research contributes to CSIRO’s work in Trusted Agrifood Exports: boosting earnings of Australian grown food.CSIRO is building and tools and technologies designed to help grow Australian agrifood exports by $10 billion by 2030 to support our farmers and boost our economy. Alongside collaborators from industry, research and government we aim to increase market access for Australian producers, reduce the cost of compliance through automation and validate biological origin of production based on food samples.

Acknowledgement

This research is supported by the Science and Industry Endowment Fund.