At Gold Kist's poultry processing plant in Carrollton, Ga., a machine vision system developed at the Georgia Tech Research Institute (GTRI) is undergoing field-testing. If it results in the success researchers expect, it could open the door for automating many visual-inspection tasks in the industry.

The technology is called a systemic screener. Installed near the front end of the chicken-processing line, cameras look for defects such as improperly bled birds and those afflicted by systemic diseases, such as septicemia and toxemia.

Unique software and algorithms provide the intelligence for translating visual data from the system's cameras into the appropriate mechanical commands for dispensation of each chicken. Those that pass the screening proceed to the next step, while unfit chickens are quickly and automatically removed from the processing line.

"It's a vision-based, closed-loop inspection and removal system -- one of the first of its kind," said Craig Wyvill, chief of the Food Processing Technology Division in the Electro-Optics, Environment and Materials Laboratory of GTRI.

By removing unacceptable birds early in the operation, the systemic screener allows subsequent areas of the plant to have "higher utilization of the processing line," he noted.

While poultry processing is already highly automated, it still depends heavily on manual processes, many of which are visually based. Thus the field trial at Gold Kist is part of a broader effort.

"We're looking at applications that span the whole gamut of the processing operation," said Wayne Daley, senior research engineer at GTRI and head of the development team. "From beginning to end, live bird to the shrink-wrapped package, there are places where visual input is required to properly process the product. We're looking at where we can apply machine vision technology, what would be required, and how we can modify our system to run tests and see how it functions."

Wyvill added: "Even though many processes have been automated, the industry still has to leave people on line to do visual screening and to take actions accordingly. If we can automate the vision functions throughout, we think we can greatly enhance the opportunity to fully automate the operation."

Even many of the inspections required by law under the authority of the U.S. Department of Agriculture (USDA) may be conducted with machine vision, Daley noted. In fact, Daley and his GTRI team started work in the area back in the mid-1980s. At the time, the Agriculture Department was sponsoring a project to augment its on-line inspectors with computer technologies.

A key focus of the project was to generate a consistent, objective and definable performance standard. When the project ended, so did the government's interest, or so it seemed.

GTRI's researchers redirected their efforts in the following years to developing a vision-based quality analysis system. Post-chill grading was the target area.

"Wayne and his crew developed a very good grading system that worked well in its ability to hold up under harsh plant conditions," Wyvill said. "The software they developed is state-of-the-art in terms of what is called 'soft' computing, which means it uses concepts like neural nets and other features to allow it to learn from its experiences and adapt."

The initial system was fairly expensive, but advances in camera and computing technology have brought the cost down considerably, Wyvill noted.

The screening system has been configured to work with USB cameras, similar to the home-computer models available for a few hundred dollars.

With the recent introduction of the Hazard Analysis and Critical Control Point Inspection Models Project, the door has been reopened to visit the USDA inspection area. The research team has seized this opportunity to transition its grading system into a systemic screening system.

Wyvill believes commercialization of both concepts is near.

"Once we get one of these systems in a plant on a commercial basis," he explained, "I'm confident we're going to see many more of these types of systems flooding out to other areas of the plant."

GTRI is already at work examining other potential applications of machine vision in poultry processing.

"For example, we envision using computer vision to determine the orientation and positioning of live products to help load them onto the shackles," Daley said.

Another important visual screening task that could be handled by computers entails identification of cosmetic defects such as tears, bruises or missing limbs. Inspectors use that information to accurately route the product to the appropriate product-handling station.

In later processing stages such as cut-up, deboning, breading and marinating, machine vision software can help make machinery more intelligent. Whereas most assembly-line applications of machine vision involve single objects of consistent size and shape, chicken parts vary considerably in those characteristics.

"For this automation to work properly, the machines need to be able to adjust for product variability," Daley explained. "You need detailed visual information that tells you about the specific part you're working on."

Carrying that point further, GTRI researchers are also working on ways to combine computer vision with X-ray imaging to improve the accuracy and thoroughness of product screening processes after deboning.

GTRI's work with poultry is attracting attention from other segments of the food industry, as well. Representatives from the citrus and bakery industries are already working with GTRI on machine vision techniques for high-speed imaging of products for quality evaluation and control of processes, Wyvill said.

"We think there's a huge potential for computer imaging as an automatic quality evaluation and control system through the whole food industry," he added. - By Jane M. Sanders

[Contact: Craig Wyvill, Wayne Daley, Jane M. Sanders]

03-Oct-2001