WEST LAFAYETTE, Ind. - Adding video-camera eyes and computer brains to conveyor belts could give consumers higher quality produce and meat. Such machine-vision systems under development in Purdue University's School of Agriculture also could increase exports of fruits, vegetables, meat and even furniture.
"The system can improve food quality when quality depends upon separation relative to exterior appearance," says Purdue agricultural engineering Professor Gary Krutz. "It could work for grading meat, pretzels, chips, peas, corn, carrots, beans and other produce."
On current sorting and packing lines, people stand elbow to elbow, watching the produce or meat cuts slide by, grabbing for blemished radishes, carrots or rib-eye steaks. Hours pass. Their eyes tire. Their minds wander. A few moldy carrots or a few fatty steaks slip past.
Machines don't tire. They don't get bored by repetition. In fact, it's what they do best. And with current technology, they can be programmed to sort by color and color patterns. They can "learn" to weed out bad produce and to sort good quality fruits, vegetables or meat cuts into uniform batches.
Krutz's prototype machine consists of a conveyor belt that runs a vegetable or a fruit or a piece of meat under a video camera attached to a computer. The computer analyzes the colors, shapes and patterns in the video image, then tells a spring-loaded arm to shunt a piece of fruit or meat into one of several piles. Krutz has run beef steaks through a prototype machine-vision separation system and has been able to sort them by percent fat and brightness of their red color. The machine could be programmed to sort by amount of marbling or bone content, he says.
"We could use it to tell consumers the percent fat in their meat," Krutz says. "And this could dramatically improve exports. At Japanese meat counters, for example, every piece is uniform in color, size and shape. You'd think they'd cloned them. We could greatly increase U.S. export markets if we could supply what the Japanese want."
Krutz says he also sees applications to the furniture industry. The same machine that sorts steaks can sort lumber by color and grain of wood. Furniture made from sorted, well-matched pieces of wood could bring U.S. manufacturers better prices and increased opportunity for sales abroad.
To correctly identify and sort a wood piece, the prototype machine needs only a sixth of a second, according to Harry Gibson, associate professor of forestry and leader of the wood products project. On a production line, that translates into 200 feet per minute, more than six times faster than human sorters. And its memory for color and grain detail exceeds that of human counterparts.
Krutz estimates the price of the hardware needed for a scaled-up version of the wood sorter at $20,000, a cost he says could be quickly recouped by increased sales.
"If an U.S. furniture manufacturer added one of these systems to a plant, they would increase their yield of better-quality products by 5 percent," Krutz says.
That would increase their ability to sell furniture in Europe and Japan, where savvy consumers demand pieces made of well-matched wood, Krutz says.
Krutz helped a U.S. seed company set up a similar machine-vision system to measure the amount of trash found in seed corn at an elevator in Toledo, Iowa.
Seed corn must be shelled (removed from the cob) after sale to seed companies, because the mechanical shellers farmers use for field corn would break seed corn kernels and kill the embryo inside. Some husks and trash are left on the unshelled corn. Seed corn companies take one or more samples from a load brought to them for sale, measure the amount of trash in the samples, then pay the farmer according to the cleanliness of his seed.
Current sampling procedures leave room for an error of as much as $2,000 per load, according to Krutz. A machine-vision system designed by Krutz and his colleagues has cut the margin of error to a few dollars by enabling seed corn buyers to quickly check an entire load, measure the trash by machine vision, and pay the producer accordingly.
ACS code/950210krutz rjg/krutz/9504ap20
Sources: Gary Krutz, (317) 494-1179; Internet, krutz@ecn.purdue.edu Harry Gibson, (317) 494-1190; Internet, gibson@ecn.purdue.edu Writer: Rebecca J. Goetz (317) 494-0461; Internet, rjg@aes.purdue.edu
