Classification of seven Iranian wheat varieties using texture features

Wheat is a major crop in Iran and plays an essential role in satisfying human nutritional needs. Depending on the geographical location as well as the climatic conditions of each region, different varieties of this crop can be cultivated. The purity of the seeds used for field cultivation has a substantial impact on both field efficiency and flour quality as the final product. To increase the seed supplied for cultivation in seed supply centers, the use of new non-destructive automatic methods for this purpose is indispensable. In this study, 7 varieties of wheat in the East Azerbaijan Province were investigated by employing a machine vision system. By extracting texture features using Gray Level Co-occurrence Matrix (GLCM), Gray Level Run Length Matrix (GLRM), and Local Binary Pattern (LBP) algorithms and classifying using principal component analysis (PCA) as the unsupervised method, along with the Support Vector Machine (SVM) and Artificial Neural Network (ANN) as supervised methods, the system accuracy was surveyed. Results indicated that modeling using a combination of all extracted features (125 features) yielded the best results by the ANN model, which the Correct Classification Rate (CCR) was obtained 100% and 95.04% for training and testing datasets, respectively. The testing dataset CCR was improved using 20 selected features by the chi-square test feature selection method up to 98.10%, which was better than the modeling results with all extracted features. Results also revealed that by using image processing and texture features extraction methods, it was possible to identify the classification of different wheat varieties with over 95% accuracy.