基于计算机视觉技术的牛肌内脂肪含量智能测定模型研究OA

To meet the demand for rapid and objective determination of beef intramuscular fat(IMF)con-tent in practical production,this study developed a computer vision-based IMF prediction model using bo-vine longissimus dorsi,also known as the eye muscle,as the study material.A total of 500 beef eye muscle samples were collected from three standardized slaughterhouses,and 1 000 cross-sectional images were ac-quired under uniform imaging conditions.IMF content was determined by the Soxhlet extraction method and used as the reference value.The Segment Anything Model(SAM)was used to automatically segment the beef eye muscle region.Fat-related image features were then extracted by combining HSV color space transformation with Otsu threshold segmentation,and the preliminary fat pixel ratio was calculated.On this basis,using the fat pixel ratio as the input variable,four regression models,including linear regression,random forest,gradient boosting decision tree,and backpropagation(BP)neural network,were constructed to compare their prediction performance for IMF content.The results showed that the preliminary fat pixel ratio was significantly positively correlated with IMF content(r=0.360 7,P ≤0.01).Comparison among different models showed that the coefficients of determination(R2)of linear regression,random forest,gradient boosting decision tree,and BP neural network on the independent test set were 0.130 1,0.398 8,0.608 5,and 0.709 2,respectively.Among these models,the BP neural network showed the best prediction performance,with a mean absolute error(MAE)of 0.025 1 and a root mean square error(RMSE)of 0.030 2,indicating good agreement between the predicted and measured values.These results indicate that the pro-posed method can achieve stable prediction of beef IMF content and provide a practical and feasible techni-cal approach for beef quality evaluation and large-scale determination of intramuscular fat content.