Preservation effects evaluated using innovative models developed by machine learning on cucumber flesh

摘要

The objective of this study was to compare the slice image features of cucumbers preserved in different ways and fresh cucumbers. The models for discrimination of spontaneous lacto-fermented, pickled using a vinegar solution and fresh cucumber slices were built based on selected textures extracted from digital images converted to color channels L, a, b, R, G, B, U, V, X, Y, Z. The average accuracies were high of up to 98 for the Random Forest and Logistic classifiers for a model built based on textures selected from a set including textures from all color channels. The discrimination accuracies for individual samples were equal to 100 for fresh cucumber (Bayes Net, Logistic and IBk classifiers), 98 for vinegar-pickled cucumber (Random Forest, Logistic), 97 for lacto-fermented cucumber (Random Forest). The values of TP Rate, Precision, PRC Area, ROC Area and F-Measure were the highest for fresh cucumber and were equal to 1.000.