With the rising concern over cholesterol, the demand for chicken now outpaces the demand for beef in the American diet. Pressured to keep pace, the poultry industry automated many of its processing techniques by the late 1970s. Inspection, however, has largely resisted automation, despite the fact that human error is often the cause of passing improperly bled or contaminated chickens through the line.Researchers at Georgia Institute of Technology in Atlanta have unveiled a machine vision system that detects defects on birds at various stages of processing, from killing and eviscerating to chilling and further processing. The systemic-screening system incorporates an off-the-shelf Sony three-chip CCD camera and a series of high-frequency fluorescent lights that distribute their output uniformly across the image plane.What makes this application unique – and more difficult than most machine vision applications – is that one unfit chicken almost always looks different from another. But, using custom software that makes use of image-processing algorithms, the new system translates and interprets the visual data from the line, identifying regions of interest in real time for feature extraction. When it recognizes a defect, the offending bird is automatically removed. Although it may not be more accurate than an attentive human sorter, said Wayne Daley, a senior research engineer at the Georgia Tech Research Institute and director of the project, the machine vision system should be more consistent. Human screeners can become distracted and allow bad birds to get by, necessitating the shutdown of the line.Up to 200 birds per minuteField tests at Gold Kist Inc.'s poultry processing plant in Carrollton, Ga., have demonstrated that the system is adept at spotting indications of septicemia and toxemia as well as poultry that has been overscalded during processing. It also recognizes birds that were improperly bled during the kill process.Commercialization of the vision system may not be far off. Work continues on fine-tuning the algorithms to keep up with the speed of real production lines, which can run as fast as 200 birds per minute. The system originally targeted birds at the front end of processing, but the researchers are modifying it to provide similar screening functions at later processing stages, such as cut-up, deboning and cooking.Daley also is part of a team working on fusing machine vision and x-ray inspection to enhance the overall accuracy of bone detection at the end of deboning operations.