Rice Variety Classification Using Next Generation Convolutional Networks

ABSTRACT Rice, a staple food for over half of the global population, exhibits remarkable diversity in grain types. However, this diversity poses significant challenges in accurate identification, often resulting in misclassification and even intentional adulteration. These issues compromise product quality and erode trust within the agricultural supply chain. Traditional identification methods, which rely on manual inspection and chemical analysis, are often imprecise, labour‐intensive, and impractical, especially when distinguishing grains based on external features such as size, shape, and texture. To address these limitations, we propose an advanced framework that leverages cutting‐edge machine‐learning techniques for rice variety classification. Our approach employs a convolutional neural network model enhanced with Transformer‐based design principles, achieving exceptional performance. The model attains an accuracy of 99.60% on one dataset and 100% on another under similar experimental conditions. Its robustness and simplicity make it a highly effective solution for practical applications. The proposed methodology holds significant promise for broader applications in agriculture and food safety, paving the way for a more transparent and reliable agricultural ecosystem that fosters trust and innovation.