Frequency and intensity texture analysis for beef quality evaluation and prediction from ultrasound images

Image analysis and ultrasound techniques, after being effectively used for live animals in our research, were used to estimate and predict intramuscular fat (present fat) for 378 hot carcasses in a packing plant in Minnesota. Image textural and statistical analysis approaches, both in the frequency spectrum domain and in the gray-tone intensity domain, were performed on the B-mode images to extract image features for tissue characterization and to develop prediction models for percent fat. A quality evaluation and prediction (QEP) software was also developed for automatic image analysis and percent fat prediction. Validation testing of these models showed very good prediction accuracy. The root mean square errors were around .93 to .96, the correlation coefficients for the predicted percent fat versus actual values were .74 to .81, and about 83 to 90 percent of the predicted residual errors were less than 1.5%. The results indicate that the percent fat prediction and the image analysis algorithms may be a valuable tool for beef quality evaluation and prediction for meat industries and livestock producers