Rice is a staple food with significant variations in quality and price among varieties. Reliable identification of visually similar types is essential for food authenticity, quality control, and fair trade, particularly in Northern Iran where minor morphological differences carry major economic implications. This study proposes a fully automated framework combining interval Type‑2 fuzzy logic for robust edge detection with a recurrent adaptive neuro‑fuzzy inference system (RANFIS) for classification. A hybrid dataset was used, comprising standardized real images and synthetically augmented samples to improve generalization under diverse illumination, noise, and background conditions. From each grain, four morphological and three color descriptors were extracted following rigorous preprocessing. Evaluations on three Iranian rice varieties—Pakistani, Hashemi Tarom, and Native Tarom—showed that the method achieved 100% accuracy on the real‑only validation set and ≥98% accuracy on the hybrid dataset, consistently outperforming SVM, KNN, and backpropagation neural networks. Training converged in under 2 minutes on standard hardware, and inference required less than 0.1 s per grain, demonstrating suitability for real‑time applications. These results confirm that integrating synthetic data augmentation with the Type‑2 fuzzy–RANFIS approach substantially enhances robustness over conventional methods, especially for challenging, visually similar varieties. The framework offers strong potential for deployment in conveyor‑based robotic inspection and high‑throughput sorting lines, enabling continuous, automated, and low‑cost quality control in modern food processing.
Sakhaei et al. (Mon,) studied this question.