Abstract: In order to establish a standardized method that can quickly and nondestructively identify the color changes in the process of mutton roasting, this study combined three algorithms (mean value algorithm, K-means algorithm and K-means+image noise reduction algorithm) based on machine vision technology to make the color recognition colourimetric card and carried out online monitoring of the color of roasted mutton. The results showed that the colorimetric cards made by the three algorithms could show the color changes in the process of mutton roasting. In order to clarify the accuracy of the three colorimetric cards, K-medoids algorithm combined with sensory experiment was used to verify the accuracy of color recognition of the colorimetric cards. The verification results of colourimetric card recognition accuracy using K-medoids algorithm showed that the accuracy of mean algorithm was 85.60%, that of K-means algorithm was 95.70%, and that of K-means algorithm+image noise reduction algorithm was 93.40%. The verification results of sensory experiments showed that the recognition accuracy of mean algorithm, K-means algorithm and K-means algorithm+image noise reduction algorithm were 67.32%, 73.71% and 68.74% respectively, the comparison showed that the color recognition accuracy of colourimetric card made by K-means algorithm was the highest for roasted mutton. The study proved that the colorimetric card can be used as the color evaluation criterion and provides guidance for barbecue meat processing. It had a good application prospect.