Inspecting poultry for safety and quality at high speed

A prototype system uses visible/near-infrared light and multispectral imaging techniques to automatically categorize the condition of chicken carcasses at industrial speeds.
07 August 2006
Yud-Ren Chen

About four decades ago, inspectors from the U.S. Department of Agriculture (USDA) began inspecting every chicken intended for sale to the general public. At that time, inspectors examined 3 billion chickens a year. Today, almost 8 billion chickens go through federally inspected processing plants annually. With this rising demand, processing lines must run at speeds as high as 140 birds per minute (bpm). An inspector's visual and manual inspection speed, though, peaks at 35bpm. Consequently, some of the research programs of the USDA Agricultural Research Service (ARS) seek new technology that will improve inspection speed, minimize problems of human error and variability, and improve the effectiveness of safety-inspection programs.

As early as 1993, my colleagues and I at the ARS Instrumentation and Sensing Laboratory (ISL) showed that visible/near-infrared (Vis/NIR) spectroscopy techniques can accurately sort carcasses according to three classes: wholesome, septicemia and toxemia (blood-related diseases), or cadaver (caused by improper bleeding). Working closely with the USDA Food Safety and Inspection Service, we have been developing automated vision systems for inspecting chickens on processing lines.

Visible and near-infrared light reflected from a carcass carries information about color, surface texture, and chemical constituents. For example, carcasses with diseases or defects often have a variety of visible skin and muscle changes, and these carcasses can be detected using Vis/NIR spectroscopic techniques. The first Vis/NIR system for inspecting chicken carcasses on a production line was developed in 1998. It consisted of a fiber-optic light probe, a spectrograph, a silicon receptor, a quartz-tungsten-halogen light source, an industrial computer, and software developed at ISL. The Vis/NIR light is transmitted through the fiber-optic line to the probe. During trials conducted at 70bpm during an eight-day period in a chicken-processing plant in New Holland, PA, our Vis/NIR system classified the carcasses—wholesome or unwholesome—with a success rate of 95% (see Figure 1).1


Figure 1. This visible/near-infrared -spectroscopic system inspects the quality of poultry.
 

In 1999, my team also developed a dual-camera system, which consists of two cameras with interference filters at 540 and 740nm, illumination, and software modules for image acquisition, control, and data analysis. Simultaneous imaging at multiple wavelengths enables detection of problems such as undersized birds or those with diseases. Trials of this system at a speed of 70bpm in a poultry-processing plant resulted in classification accuracies of 94 and 87% for wholesome and unwholesome carcasses, respectively.2 Moreover, the accuracy of the Vis/NIR spectrophotometer and multispectral imaging systems was similar in processing-line trials and previous laboratory studies. Thus, these systems show promise for separating unwholesome and wholesome carcasses on poultry-processing lines.

Most recently, we upgraded the Vis/NIR subsystem with a CCD spectral detection sensor, object-oriented program modules, and hardware-control instruments. This system was successfully tested in a commercial chicken-processing plant at speeds of 140–180bpm on the kill lines.3 On the kill line, birds are stunned, bled, scalded, defeathered, and have their heads and feet removed, before they are put into the evisceration lines, for which the processing speed is reduced to 60–90bpm. In addition, we improved the multispectral imaging system by using a common aperture-imaging system with a newly available electron-multiplying CCD to overcome the challenges of low-light conditions and exposure-time restrictions imposed by high-speed operations. This allowed multispectral imaging with long-life LEDs that minimize lighting variations (see Figure 2). This system can detect unwholesome carcasses at speeds higher than 140bpm.


Figure 2 An electron-multiplying CCD minimizes variations in light in the multispectral imaging system.
 

The ISL's automated poultry-inspection systems are now industrial prototypes that can operate at processing-line speeds as high as 180bpm. Through a cooperative research and development agreement, ISL is working with a major manufacturer to commercialize the inspection system.


Author
Yud-Ren Chen
Instrumentation and Sensing Laboratory, USDA Agricultural Research Service
Beltsville, Maryland

Recent News
PREMIUM CONTENT
Sign in to read the full article
Create a free SPIE account to get access to
premium articles and original research