Project looks at how to modernize meat inspection

New technology could bring benefits to meat inspection in the United Kingdom but there are still issues to overcome, according to a report.

A project assessed the feasibility of using sensor technologies and advanced data analytics for poultry inspection. It focused on post-mortem inspection and included technologies such as visual, near-infrared, infrared and hyperspectral, X-ray and ultrasonic as well as IT-enabled benefits.

Poultry is the top consumed meat in the UK. Inspection is manual and challenging because of the short time to check each bird and the constant level of concentration required. Human error is possible and it is subjective based on opinion and experience of the meat inspector.

The project team visited a food facility in November 2019. This site uses optical imaging and X-ray in primary poultry processing. These technologies are to detect hock burn, to grade the carcass or to find foreign bodies in final products. The factory slaughters 250,000 birds a day and has two full-time meat inspectors working on each of their two processing lines.

Replace or complement existing methods
The data capture method for rejections is currently done manually using clipboards and requires repeated data entry, according to the report published by the Food Standards Agency (FSA).

Five IT points identified included any system must be easy to use, reliable, robust, be well-understood and needs to solve a problem.

Concerns were raised based on previous failed attempts to install new technology, which added complexity for the inspector. A fear from meat inspectors was that the technology must be at least as reliable as the person, many of whom have decades of experience. Technologies could be used alongside the current inspector role to reduce the workforce burden.

Plants in which these technologies are being installed are regularly cleaned using corrosive chemicals and large quantities of water. Depending on where it is fitted, there may be variations in environmental conditions such as humidity and temperature.

Poultry processors operate in a competitive sector with tight margins. Any changes to timings or stopping the production line can mean the loss of vast quantities of product.

Other factors include concerns about job security, the impact on day-to-day operations, suitable training and an understanding of what data is needed and how it might inform actions.

Potential drawbacks and benefits
Seven problems were identified such as limitation of technology, the multiple stakeholders involved, retailer influence, legal issues and a lack of time for experimentation.

Despite technology, a meat inspector would be needed to identify other conditions or to remove birds from the line and place them in the correct category bin. Automating this process would require a major investment in factory redesign. Regulator, food firm, retailer and consumer needs, concerns, and drivers do not always align, according to the research.

There are likely to be legal issues around who has responsibility for food safety, technology, and the data and changes to roles would require legislation to be changed.

Artificial intelligence, sensor and data analytic technologies would need to be tested before implementation to minimize disruption to the food firm. There are multiple stakeholders involved in the production process, making things complex. This may present challenges for achieving agreed standards for data access and governance, responsibility and liability for automated decisions, and committing financial resources.

Business benefits of new technology would include improved data usage and reliable measurements in real-time. Data could be shared with farms to allow earlier identification of issues and help maintain retailer trust.

Automation of the inspection process may reduce human error where healthy birds are incorrectly rejected or decrease the instances of retailers refusing products where poor quality birds have been missed. Improving accuracy of inspection will save time and money.

The project found deep learning could identify abnormal color from carcass images with a sufficient number of training images, but more efficient data labeling methods are required. However, some results from models would lead to acceptable carcasses being classed as unacceptable.

Imaging and classification could be an initial screen to identify carcasses where a condition may be present and these could be highlighted to an inspector for a more detailed look and decision.

Hyperspectral optical and X-ray imaging methods were also able to identify quality issues such as wooden breast and white stripe in chicken breasts but cost could be a barrier to adoption.

(To sign up for a free subscription to Food Safety News, click here.)