Classification of Iranian king of spices (saffron) types by texture features

Precise authentication of saffron types in Iran, as the main producer not only prevents economic losses for producers and consumers, but also minimizes the threat of adulteration. Therefore, in the present research, combination of machine vision method and texture feature extraction algorithms was applied to classify 3 main types of Iran commercial saffron including Sargol, Negin, and Pushal. Three texture feature extraction algorithms including Local Binary Pattern (LBP), Gray Level Co-occurrence Matrix (GLCM), and Gray Level Run Length Matrix (GLRM) and combination of them were employed. Various machine learning models including Discriminant Analysis (DA), Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Artificial Neural Network (ANN) models were applied to evaluate the results of classification. SVM model with Linear kernel and all features had the best outcome with 87.22% for test dataset. Furthermore, four feature selection algorithms including Chi-Square Test (CST), Minimum Redundancy Maximum Relevance (MRMR), ReliefF, and Kruskal Wallis were used to select the most important features. Selected features by CST algorithm with SVM model had the best outcome with the accuracy of 76.7%. The results of present research confirm the applicability of machine vision technique for classification of commercial saffron types which is significant in saffron industry. • Integrated machine vision and multi-texture feature extraction (LBP, GLCM, GLRM) for saffron authenticity verification. • 87.22% classification accuracy using SVM (Linear kernel) for distinguishing three Iranian commercial saffron types. • Applied feature selection algorithms to optimize classification performance and reduce computational complexity for industry use.