Prediction of lamb tenderness using image surface texture features Original Research Article

This paper investigates the usefulness of raw meat surface characteristics (geometric and texture) in predicting cooked meat tenderness. Twelve geometric features were measured from each of the acquired lamb chop images. In addition, 136 texture features including 36 difference histogram, 90 co-occurrence and 10 run length texture features were also extracted. Four feature sets comprising six geometric, four difference histogram, eight co-occurrence and four run length features were generated based on the results of dimensionality reduction. These four feature sets, individually and in different combinations, were utilised to predict cooked meat tenderness using neural network, linear and non-linear regression analyses. Non-linear regression analysis produced higher coefficients of determination (R2) than linear regression analysis. The neural network analysis produced highest R2 of 0.746 using 14 (geometric and co-occurrence) features. The non-linear regression analysis produced highest R2 of 0.602 using 22 (geometric, co-occurrence, difference histogram and run length) features. This study shows the potential of texture analysis, in combination with image analysis, for prediction of meat tenderness.