Innovative Computer Vision System Provides Portion and Process Control
for Further Processed Product Lines
Responding to increased demand for ready-to-cook and fully cooked
products, poultry processors operate further processing lines that
produce a myriad of chicken products, such as the popular breaded breast
fillet or nugget. Millions of pounds of these products are produced
each week, all of which have a stringent set of customer specifications.
For example, some customers require that fillets not deviate from a
pre-determined size and shape.
|
Parker McGee (left), a co-op student, and Colin Usher (right),
a research scientist, perform in-plant tests on the overline
screening/sorting prototype system. Developed by Georgia Tech
researchers, the innovative computer vision system screens individual
meat and poultry portions on-line for both volume and visual
quality.
|
Adherence to such guidelines requires
close monitoring by plant quality control personnel. Samples are
routinely removed from the processing
line to confirm that the product is meeting the desired specifications.
However, this manual process can never guarantee that all product
shipped meets the customer’s acceptability criteria.
One way to achieve
this is through 100 percent inspection and grading. And, of course,
that means automation.
Researchers with Georgia Tech’s
Food Processing Technology Division, with funding from Georgia’s
Traditional Industries Program for Food Processing, have developed
an innovative computer vision system
for on-line screening of individual meat and poultry portions for both
volume and visual quality.
“The requirements being placed on chicken and beef producers
to meet the needs of their customers in the further processed and case-ready
products areas are challenging to say the least,” comments Wayne
Daley, associate division chief of FPTD and project director. “As
industry continues to produce more of this product mix, systems such
as ours will serve to enhance plant efficiencies and reduce costs.”
According
to Daley, the current manual process is labor intensive and does
not provide data in real-time to support decision making. An automated
system, such as the one developed by his research team, allows processors
to optimize the production process while also providing data on conformance
to customer specifications.
The prototype system screens individual
portions, on-line, for both volume and visual quality. This is used
to help ensure that the product
meets customer standards while also generating feedback data that
can be used to optimize process operations. It employs unique lighting
and imaging features that monitor key product characteristics such
as size, weight, shape, height, and surface appearance for defects
like bruising, tears, and fat coverage, along with a host of other
customer specifications. In fact, researchers believe it is the only
system they are aware of that is able to sort product not only on
weight
and dimensions but also on surface quality defects.
Working with industrial
partner, Wayne Farms, the research team performed a series of performance
tests at its processing plant in College Park,
Ga. The system’s design accommodates a parts rate of 100 pieces
per minute.
The product’s height is determined by a laser-based
structured lighting system, which also drives the acquisition of a
visible image
of the product. This image is then processed to assess whether or not
the product meets the desired specifications. The generated data then
passes to a grader that determines which grade to give the product
and sends that grade to the sorter which then sorts the product.
During
testing, the system demonstrated the ability to monitor the process
and to provide real-time feedback and guidance to operators.
Such information, explains Daley, allows for fast process adjustments,
and if sorting is done based on these parameters, it could be used
to guide rework.
A provisional patent has been filed on the use of a
dynamic lighting system to provide flexibility in sensing and image
acquisition. In
addition, discussions are now underway with Gainco, Inc., in regard
to licensing the technology.
Daley says in the future, one issue the
team plans to address is the loading of the machine. The system is
currently manually loaded to
obtain the desired presentation for accurate assessment of shape
and size.
Additionally, says Daley, the sensing and decisions made
by the system lay a solid foundation for developing the next generation
of second
and further processing machines, “smart machines,” that
react to changes in product and processing mix.
|
 The system employs unique lighting and imaging features that
monitor key product characteristics such as size, weight, shape,
height, and surface appearance for defects like bruising, tears,
and fat coverage, along with a host of other customer specifications.
The images above show a butterfly fillet that has been processed
by the system, highlighting areas of fat coverage.
|
|
