In a research project sponsored by the US Poultry & Egg Association, Douglas P. Smith of North Carolina State University and Kurt C. Lawrence, Bosoon Park and William R. Windham of the USDA-ARS Russell Research Center in Athens, Georgia, looked at the detection of bone and physical contamination of boneless broiler meat.

They explain that bone fragments embedded in boneless chicken meat are significant food hazards to consumers. Commercial poultry processors may lose customers and spend considerable resources each year for insurance claims and legal fees due to broken bones found in de-boned chicken meat. The clavicle (wishbone), which is a long, sharp and completely calcified bone, is the physical hazard predominantly found in boneless and skinless chicken fillets.

The USDA Food Safety and Inspection Service (FSIS) in its 1995 Public Health Hazard Analysis Board concluded that bone particles less than 1cm are not a safety hazard, particles between 1 and 2cm are a low risk, and particles greater than 2cm have the potential to be a safety hazard and may cause injury.

X-ray imaging is currently the primary method for detecting bones in poultry meat, say the researchers, but there is a need to improve the detection accuracy and to explore alternative methods for bone detection.

The objectives of this project were: 1) to develop an accurate detection system for bones, bone fragments, and other physical hazards in or on boneless broiler meat using hyperspectral (HS) imaging and 2) based on results from the hyperspectral imaging system, a multispectral (MS) imaging system was to be used to detect bones and hazards at real time processing conveyor speeds.

A system to detect bones and bone fragments in boneless skinless breast fillets was successfuly developed. Optical imaging of chicken fillets was dominated by multiple scattering properties of the fillets. Resulting images from multiple scattering were diffused, scattered and had low contrast. A fusion of hyperspectral transmittance and reflectance imaging, which is a non-ionized and non-destructive imaging modality, was investigated on fillets compressed to a uniform thickness of 1cm.

An image formation model, called an illumination-transmittance model, was applied for correcting non-uniform illumination effects so that embedded bones were more easily detected by a simple segmentation method with a single threshold value. Predicted bones from the segmentation were classified by a nearest-neighbour classifier that was trained from a spectral library of mean reflectance data of chicken parts including, meat, fat, connective tissue and embedded bones.

Experimental results with chicken breast fillets and bone fragments resulted in a detection accuracy of 100 per cent for bones greater than 2cm in length. However, the false-positive rate was 10 per cent, primarily from thick fatty material. Attempts to detect embedded bones in thigh meat were less successful.

To simulate a multispectral imaging system that could operate in real time, it was observed that between 550 and 650nm, fat had a higher transmittance spectra. Thus, for the multispectral imaging simulation, a wavelength at 608nm was used to detect bones. Using the single wavelength data with structured light ensured that a system could easily be developed to rapidly image the fillets in real time. However, since the false positive rate was not reduced below 10 per cent, the results were not good enough to warrant developing a stand-alone multispectral imaging system.

Since whole fillets from smaller birds are widely used in the fast-food industry to meet customer requirements, the imaging system could be used to screen skinless breast fillets for bones. Furthermore, most fast-food companies require that the whole fillet, with batter and breading applied, be no thicker than 2.5cm, which means approximately 2cm maximum thickness for the fillet. This thickness can be temporarily compressed to 1cm without loss of structural integrity, as is typically done en masse with a conveyor belt flattener. However, in the short term, improvements in the false positive rate are needed for implementation.

December 2011