Innovative Computer Vision System Detects Foreign Material on Food Processing Lines

John Stewart, senior research engineer, is spearheading efforts to build a color vision system that will automatically detect and then remove colored foreign objects from the food stream.

Plastic is commonly used by food processing operations in liners for containers, disposable gloves, food testing instruments, hearing protection, identity badges, conveyor parts, and guides along conveyors. Despite extensive safeguards, these items or small fragments of them sometimes make their way into the product stream and end up in the finished product. Unfortunately, there is no easy way to find plastics once they enter the high volume flow of a commercial food production line. However, researchers with Georgia Tech’s Agricultural Technology Research Program are busy working to find a solution. They believe an answer might lie in computer vision technology. Using the technology along with sophisticated software algorithms, the team has developed an overline color vision system that is proving viable at detecting plastic fragments that have become lodged in finished product.

The squared segments indicate that the system’s software has detected a foreign object (in this case, plastic glove pieces) in the sample product.

To help detect foreign material in products on food processing lines, most producers currently use plastic items with colors that stand out from the product stream in hopes that an employee will find the items, explains John Stewart, senior research engineer. “The goal of our research is to build a color vision system that will automatically detect and then remove these colored objects from the food stream.”

“The detection of foreign objects and contaminants in food is a critical safety task. With the amount of automation present at every level of food production and the rates at which food is being produced, it is becoming increasingly more important to have systems that can automatically detect foreign matter along the way,” adds Stewart.

“A computer vision system is much more effective at this task than human observers. First, the vision system continuously watches the product stream and does not become distracted or daydream like a human observer. Second, the vision system can freeze motion on a relatively high-speed belt, and its resolution can be specifically tailored for the observation task at hand,” says Stewart.

According to Stewart, these capabilities allow the system to operate on a high-speed line that can separate the product and present more surface area for screening. And, Stewart says, by automating the detection process, there is a digital record of any foreign objects detected that can aid in identifying the point where the material entered the process.

Stewart and his research team have produced a working prototype. The system uses a blue color scheme to detect foreign objects. The system concept is simple, he explains. An overline vision system is installed at an appropriate point in the process, typically next to the final metal detection cell. The vision system is trained using color discrimination algorithms by passing unadulterated product under it. The information on the product appearance is then stored in a product dependent profile, which can be called up remotely as the product mix on the line changes. Once the training mode is complete, the vision system looks for any object that does not match the product profile. If the system detects a problem, it sounds an alarm, saves a picture of the problem product, and activates a product kickoff device. If the system is placed next to a metal detector, it is possible the two systems can share the same kickoff device.

In preliminary tests, the system has demonstrated a near 100% detection rate for blue and green glove parts larger than 1.5 mm. The system operates at up to 12 feet per second while visually inspecting all products at least twice. The system has also proved successful in detecting green glass, zip ties, and larger pieces of blue tinted plastic box liner.

Stewart says the team is currently focusing its efforts on building a production-scale prototype. In fact, the team was recently awarded a grant from Georgia’s Traditional Industries Program for Food Processing to fund the effort. The team plans to work with Gainco, Inc., an equipment manufacturer based in Gainesville, Ga., to produce the prototype. The team has also partnered with Wayne Farms LLC to conduct field studies at the processor’s Oakwood, Ga., processing plant.