Vision-Based Quality Grading of Beef Steaks Using Marbling Distribution Analysis and Lean Meat Color Classification

Area of Science:

• Food Science and Technology
• Computer Vision
• Image Processing

Background:

• Automated quality grading of beef steaks is crucial for the food industry.
• Surface frost on frozen beef products creates specular reflections, hindering accurate segmentation of fat and lean tissues.
• Existing methods struggle with illumination variations and complex texture analysis in beef quality assessment.

Purpose of the Study:

• To develop a robust vision-based framework for automated inspection and quality grading of beef steaks.
• To address the challenge of frost-induced artifacts in frozen beef image analysis.
• To integrate fat marbling distribution and lean-meat color evaluation for comprehensive quality assessment.

Main Methods:

• Applied homomorphic filtering to mitigate frost-induced illumination artifacts.
• Utilized curvelet transform and square-ring filtering for multi-scale fat-lean segmentation.
• Extracted marbling features (convex hull, skeleton) and performed chi-square tests for distribution analysis.
• Employed Support Vector Machine (SVM) for lean-meat color classification based on RGB features.
• Integrated marbling and color data using a weighted grading strategy.

Main Results:

• Achieved high accuracy in fat segmentation: 92.68% detection rate, 4.97% false-positive rate, and 94.09% correct classification.
• The SVM-based lean-meat color classifier demonstrated 96.67% accuracy.
• The integrated grading framework reached an overall accuracy of 90.38%, showing strong agreement with human evaluations.

Conclusions:

• The proposed vision-based framework effectively automates beef steak inspection and quality grading.
• The integrated approach successfully overcomes challenges posed by frost artifacts, enabling precise marbling and color analysis.
• The system demonstrates significant potential for objective and consistent beef quality assessment in the industry.