Development of a convolutional neural networks‑based model to classify the rice varieties

Abstract

Rice (Oryza sativa L.) is an important staple food in India, particularly in the North Eastern Hill (NEH) region. The surge in demand for rice, along with its significance in international trade, highlights the need for accurate identification of rice varieties. To address this, we constructed a dataset having four rice varieties— Bhalum-5, Shahsarang, Nagina-22, and IR-64. Our dataset comprises high-quality images of rice seeds, captured with smartphones having different seed counts, including 1-seeded, 5-seeded, and 10-seeded per image. A novel classifier was developed for the classification of rice. Using Convolutional Neural Networks (CNNs) with an architecture comprising 5 layers, the developed model demonstrates significant efficacy in accurately categorizing rice seeds. Experimental results revealed significant achievement and a maximum classification rate of 91.0% for the prominent rice variety (Shahsarang) cultivated in the region. This outstanding accuracy of the developed CNN-based classifier emphasizes its potential applicability in the identification of rice varieties. Thus, this research addresses the pressing need for reliable classification methods for the identification of rice varieties.