Utilizing machine vision in wheat seed classification is an impartial method which can help in online verification and increases the accuracy of seeds selection in real applications of agricultural industry. In this study, the efficiency of using some texture features to classify nine Iranian wheat seed varieties was evaluated. 1080 grayscale images of bulk wheat seeds (120 images of each variety) were acquired with similar illumination condition (florescent ring light). 77 textural features were extracted from LBP (Local Binary Patterns), LSP (Local Similarity Patterns) and LSN (Local Similarity Numbers) matrixes. In order to select the most significant textural features, Stepwise discrimination method was individually used for each matrix and also for all matrixes. With this method features were ranked based on their level of contribution in classification when they are evaluated individually and all together. LDA (Linear Discriminate Analysis) classifier was employed for classification using top selected features. The average accuracies of the classifier were 63.89%, 77.04% and 75.74% when top 10 features of LBP, LSP and LSN matrices were used respectively. The average classification accuracy was 90.37% when top 30 features selected from all feature groups were used in the classifier.

Keywords