On-line Ham Grading using pattern recognition models based on available data in commercial pig slaughterhouses

Area of Science:

• Agricultural Science
• Food Science
• Machine Learning

Background:

• Subcutaneous fat thickness in hams is critical for dry-curing and final product quality.
• Manual measurement in slaughterhouses is standard but can be time-consuming and subjective.

Purpose of the Study:

• To develop an automated ham classification method simulating manual fat thickness assessment.
• To utilize carcass data and intrinsic pig information for objective classification.

Main Methods:

• Evaluated decision tree, SVM, k-nearest neighbour, and discriminant analysis algorithms.
• Classified 4000 hams into thin, standard, semi-fat, and fat categories based on fat thickness (0-20+ mm).
• Used data from carcass automatic classification equipment (AutoFom) and pig attributes (sex, breed, weight).

Main Results:

• Support Vector Machines (SVM) with a Gaussian kernel demonstrated the highest reliability.
• The SVM model achieved a 73% success rate in classifying ham fat thickness.
• The model integrated carcass and intrinsic pig data for comprehensive classification.

Conclusions:

• The proposed automated classification method is a viable online tool for slaughterhouses.
• SVM offers a reliable approach for objective ham fat thickness classification.
• Automated classification can enhance efficiency and consistency in ham processing.