Predicting pork color scores using machine vision and support vector machine technologies

Abstract Text: The objective was to investigate the ability of image color features to predict subjective pork color scores (NPB, 2011). Enhanced retail pork center-cut loin chops in overwrapped packages were purchased and transported to North Dakota State University for analysis. Chops were assessed on the cross-sectional surface for subjective color score by one trained observer (score 2, n=75; score 3, n=284; score 4, n=240; score 5, n= 86) and instrumental color (CIE L*, a*, and b*) using a Minolta Colorimeter (CR-410, 50 mm diameter orifice, 2° observer, C light source; Minolta Company, Ramsey, NJ). Images of pork loin samples were then acquired using a machine vision system which include a charge-coupled device (CCD) camera (MV-VS141FM/C, Micro-vision Ltd., China) with a 5mm C-mount lens (aperture of f/1.4 to 16C, H0514-MP2 1/2" fixed Lens, computer CBC Americas Corp, USA). After background and muscle segmentation from images, 18 image color features (mean and standard deviation of R, G, B, H, S, I, L*, a*, b*) were extracted from three different color spaces: RGB (Red, Green, Blue), HSI (Hue, Saturation, Intensity), and L*a*b*. Color features were submitted to partial least squares (PLS) regression and support vector machine (SVM) regression analyses to establish prediction models for different color scores. A subsample (80%) of data were used to train the PLS and SVM models, which were then validated on the remaining 20% of the data. For color score 2, the accuracy of the PLS model was 86.7% classified correctly and the SVM model was 93.3% classified correctly. For color score 3, the PLS model predicted 81.5% correctly and the SVM model predicted 79.6% correctly. For color score 4, PLS model predicted 89.6% correctly and the SVM model predicted 91.7% correctly. For color score 5, the SVM model predicted the highest accuracy of 72.7% while the PLS model’s prediction accuracy was 90.9%. Image color features isolated through the development of PLS and SVM models show potential as a means to predict pork color scores.

Keywords: machine vision, pork, color score