Skip to main content
ARS Home » Research » Publications at this Location » Publication #240861

Title: Automatic inspection using machine vision for food safety

Author
item Yang, Chun Chieh
item Chao, Kuanglin - Kevin Chao
item Kim, Moon

Submitted to: Computers in Agriculture
Publication Type: Proceedings
Publication Acceptance Date: 6/10/2009
Publication Date: 7/22/2009
Citation: Yang, C., Chao, K., Kim, M.S. 2009. Automatic inspection using machine vision for food safety. In: Proceedings of the 7th Congress of Computers in Agriculture and Natural Resources, June 22-24, 2009, Reno, Nevada. Paper No. 09-7549.

Interpretive Summary: A machine vision system was developed for automated poultry carcass inspection. The use of this automated system will significantly benefit poultry processing plants by identifying unwholesome birds for removal from the processing line to minimize food safety risks, increase production efficiency, and reduce labor and cost requirements. The line-scan machine vision system scans the surfaces of poultry carcasses hung on a high-speed processing line, automatically detecting each individual carcass as it enters the field of view for inspection, and immediately identifying carcasses that are unwholesome. During in-plant testing, the line-scan machine vision system successfully inspected over 60,000 carcasses on a poultry processing line operating at a speed of 140 birds per minute.

Technical Abstract: An automated machine vision system using line-scan spectral imaging and a multispectral differentiation algorithm was developed for differentiation of systemically diseased and wholesome chicken carcasses. During line-scan imaging of chickens on a high-speed processing line, the automated inspection system located the region of interest (ROI) for each chicken and used the relative reflectance intensity at 580 nm and the ratio of intensities at 580 and 620 nm from ROI pixels as inputs to the differentiation algorithm. To locate the ROI on each chicken, the system first located the starting point on the leading edge of the bird, then located pixels between the 40 percent and 60 percent boundaries in each line scan until the last line scan containing relevant ROI pixels was found. Two mapping functions converted the inputs to the decision output, which represented the possibility for the carcass to be systemically diseased. When the output was greater than or equal to the decision threshold value of 0.60, a carcass was identified as being systemically diseased. Hypesrpectral imaging in March 2007 for algorithm development and real-time online multispectral inspection in July 2007 was performed on a commercial poultry processing line operating at 140 birds per minute (bpm). Online system performance was evaluated by comparison of inspection system decisions with bird-by-bird identifications by a USDA veterinarian. All veterinarian-identified systemically diseased birds were correctly classified by the inspection system. Comparison of the system’s overall count of wholesome and systemically diseased birds with the overall counts by human inspectors was also favorable. The results showed that the system could successfully differentiate wholesome and systemically diseased carcasses on a 140 bpm processing line.