Study on methods for measuring beef color and predicting storage time based on computer vision

In this study, a non-destructive and accurate method based on computer vision and machine learning was developed to measure beef surface color and to predict storage time and the relative proportion of oxymyoglobin. Images of the Longissimus thoracis (LT) muscle were acquired with a camera, and the red muscle region was segmented by integrating GrabCut and Otsu binarization techniques to extract RGB values, which were subsequently converted to CIE L*, a* and b* color space values. Colorimetric characteristics derived from computer vision were compared with those obtained from a traditional colorimeter, and the results revealed higher sensitivity to temporal changes and can provide a more representative reflection of the actual appearance of beef. A convolutional neural network (CNN) model trained by images derived from computer vision achieved R² values of 0.926 and 0.893 for storage time and oxymyoglobin content prediction, respectively, thereby providing an efficient and scientifically robust framework for assessing beef freshness.