Identification of three Iranian Rice Grain Varieties in Mixed Bulks Using Different Textural Features and LVQ Neural Network

Due to variation in economic value of different rice varieties, reports indicate the possibility of mixing different varieties on the market. Applying machine vision techniques to classify rice varieties is a method which can increase the accuracy of classification process in real world applications. In this study, several textural feature groups of rice grains’ images were examined to evaluate their efficacy in identification of three Iranian rice varieties (Tarom, Fajr, Shiroodi) in the mixed samples of these three varieties. On the whole, 666 images of rice grains (222 images of each variety) were acquired at a stable illumination condition and totally, 41 textural features were extracted from different matrices of grain images. Fisher's coefficient method, Principal Component Analysis method and a combination of these two methods were employed to rank and select the most significant features for the classification. The so called LVQ4 neural network classifier was employed for classification using top selected features. The classification accuracy of 97.96, 100 and 97.83 percent by using gray level matrix, 96.23, 100 and 100 percent by using co-occurrence matrix, 100, 97.50 and 100 percent by using local binary pattern matrix, and 100, 97.67 and 100 percent by using the whole textural features were obtained for Fajr, Tarom and Shiroodi, respectively. These results indicate that image processing can be a suitable tool for identification of different rice varieties. Although using the features of all matrices leads to less classification error, the features of each matrix also provides reasonable accuracy.