Abstract:

In order to establish a method to evaluate snowflakes beef marbling grades, the main factors affecting grading
marbling were identified by comparing the image features with artificial rating criteria of different snowflakes beef
marbling grades. This study presented the geometric feature parameters, geometric distribution feature parameters and
statistical feature parameters affecting marbling grade. The geometric feature parameters mainly reflected the marbling
area, perimeter and so on. The geometric distribution feature parameters mainly reflected the different deposition densities of
large, medium and small fat particles in the marbling image. The statistical feature parameters mainly reflected the marbling
abundance and marbling distribution uniformity. Correlation analysis between the features parameters extracted and
snowflake beef marbling grades was conducted. Decision tree models were established based on C4.5 and CART algorithm,
and the results showed that the prediction accuracy of three-level and five-level grades were 91.80% and 92.31%, respectively,
however, the model for the four-level sample model was invalid and the misjudgment results were mostly three-level. The
same problem existed in the prediction accuracy of models based on CART algorithm.

Key words: snowflake beef, marbling, classification model, decision tree