W.VAW.2202 - Animal Welfare Ecosystem

Summary

The cattle livestock export market to Vietnam has animal welfare compliance risks that have resulted in increased regulatory oversight and costs to stakeholders. The Animal Welfare Ecosystem (the Ecosystem) project tested if bringing aligned stakeholders into a more collaborative environment would increase adoption of innovative technologies and reduce the cost of regulation. Facial verification was initially proposed in the LiveCorp MRAG report in Oct 2019 as a means to add an extra layer of validation within traceability systems. Accurate
traceability records and suitable systems are regulated under the Australian Exporter Supply Chain Assurance System (ESCAS).

The project had two components a) the research element b) commercial implementation and
testing.

The research element of the project utilised both deep learning NN-based methodology alongside a metrics-based approach. Initially, 5,800 frontal images of cattle were developed using deep learning-based methodology for image segmentation. Then the metrics-based approach was applied using the geometry of the eye and muzzle locations. This was able to demonstrate the combination can be used for cattle identification with the accuracy of 95% (83%) when the cattle in the new image was predicted to be different from (same as) the claimed one. For the NN-based approach a preliminary architecture of the network combined with the triplet loss function demonstrated an accuracy of 85% (86%) when the cattle in the new image was the different from (same as) the cattle with the claimed RFID.

The project demonstrated the limitations of integration or reliance on commercial systems for early-stage research. Images from Vietnam that were already being captured were not consistent or high enough quality and so more controlled data was collected for this project.

While the end goal of autonomous facial verification is possible it is imperative that further research and tests progress from well-controlled conditions to more complicated conditions, such as lower illumination, videos of walking cattle, videos captured via a fixed camera. While the ecosystem provides a future pathway for commercial implementation, the requirement for the controlled research component cannot be bypassed.

Key findings

Trialed remote auditing - at this stage connectivity issues and user issues mean this technology is not a replacement for physical auditors in Vietnam however it can be used as a means of audit verification.

Facial verification technology - developed an algorithm. Scan rates on moving images in raceways was 60%, journal article developed and algorithm was able to determine whether an image was the same as or different from another animal at a rate of 95%.
Animal welfare training - 50% of supply chains in Vietnam received training.

Benefits to industry

Facial verification technology tested - this was on the recommendation of an independent review in to traceablity systems in Vietnam which recommended the use of this technology in ensuring ESCAS compliance - the technology currently is unviable and not accurate enough. This projected was able to demonstrate this and advise industry in order to prevent costly barriers to trade being proposed from regulators.

Remote auditing - demonstrated the best use case for remote auditing technology (as it currently stands) - verification rather than replacement of auditors.

Animal welfare training -reduced risk of ESCAS non compliance through upskilling and refreshing understanding.

MLA action

MLA continues to work in the animal welfare space, but without a formal ecosystem in place. No further work is being looked at in remote auditing or facial verification due to the cost and limitations of these technologies.

Future research

Ecosystem worked well by providing a platform for exporters to work together to serve common goals and to scale up adoption or trial of innovations. Suggest for future attempts at in-market adoption, a similar working group formed. Ecosystem has since lapsed as the cost of maintaining an ecosystem is only justifiable when working towards certain goals.