Predicting beef tenderness using color and multispectral image texture features.

Area of Science:

• Food Science
• Meat Science
• Agricultural Engineering

Background:

• Beef tenderness is a critical quality attribute influencing consumer satisfaction.
• Objective prediction of beef tenderness is challenging due to variability in muscle characteristics.
• Current methods for assessing beef tenderness can be destructive or subjective.

Purpose of the Study:

• To evaluate the efficacy of raw beef surface texture characteristics in predicting cooked beef tenderness.
• To develop and validate predictive models for beef tenderness using image analysis.
• To compare the performance of different analytical methods (STEPWISE regression and Support Vector Machine) for tenderness prediction.

Main Methods:

• Extraction of color and multispectral texture features from raw beef steaks using two imaging systems.
• Classification of steaks into 'tough' and 'tender' groups based on Warner-Bratzler shear force measurements.
• Development of prediction models using STEPWISE multiple regression and Support Vector Machine (SVM) algorithms.
• Training and validation of models using 80% and 20% subsets of the data, respectively.

Main Results:

• The Support Vector Machine (SVM) model achieved 100% accuracy in identifying tender steaks from color images.
• The STEPWISE regression model correctly identified 94.9% of tender steaks using color image data.
• For multispectral images, the SVM model demonstrated an average accuracy of 91%, while STEPWISE achieved 87% accuracy in predicting beef tenderness.

Conclusions:

• Raw beef surface texture, particularly when analyzed using multispectral imaging, is a valuable predictor of cooked beef tenderness.
• Image-based texture analysis offers a promising non-destructive method for objective beef quality assessment.
• Support Vector Machine (SVM) models show high potential for accurately predicting beef tenderness compared to traditional regression methods.