Classifying scratches on chicken carcasses on processing line using two-dimensional image processing technique

Abstract

Importance of the work: In Thailand, poultry processing has increased steadily. However, the selling price per carcass has dropped due to quality issues.
Objectives: To explore the possibility of detecting scratches on chicken skin during processing of carcasses.
Materials & Methods: A sample of 100 chickens was analyzed using photographs taken of each side for leg quarters with backbone on the midsagittal plane to categorize meat quality based on input from a specialist in chicken damage classification. An algorithm was developed to analyze each image based on sizing objects in pixels and the light intensity of extracted features.
Results: The first test resulted in an overall accuracy of detecting scratches of 68%. The main issue was that the program incorrectly identified shadows on the carcass as scratches. After adjustments to reduce errors from shadowing, the second test resulted in the algorithm correctly identifying 94% of the scratches. However, for certain samples, the algorithm was inconsistent with the quality level classification by the specialist because the specialist also considered the texture of the chicken skin during visual inspection.
Main finding: Using two identical cameras and light bulbs is recommended to capture better photographs and to enable more accurate in-depth analysis, supported by statistical data. The current results met the functionality requirements and should be developed further to assist automated quality inspection techniques in the future.