Book Title: Multidisciplinary Research Nexus: Exploring Intersections of Knowledge
Editor: Prof. Amos R
Chapter: 3
DOI: https://doi.org/10.59646/559/3
Authors: Yashaswini S, and Amos R
Abstract
Accurate and automated classification of rice grain varieties is important for quality control, authentication, and supply-chain traceability. This paper surveys recent deep-learning approaches and proposes a reproducible framework using deep convolutional neural networks (CNNs) and vision-transformer (ViT) variants for classification of rice grain varieties from images. We describe dataset preparation (public rice image datasets), standardized preprocessing and augmentation, network architectures (transfer learning with EfficientNet/ResNet, custom CNN, and ViT), and evaluation metrics (accuracy, precision, recall, F1, confusion matrices). We summarize performance achieved in recent literature and provide recommended experimental configurations for researchers and practitioners. Finally, we discuss limitations—dataset bias, imaging conditions, and real-world deployment—and outline directions for multimodal and hyperspectral methods to improve robustness.
Keywords: Rice grain classification, Deep neural networks, Convolutional neural networks (CNN), Vision Transformers (ViT), Image-based varietal identification, Agricultural automation, Computer vision, Machine learning in agriculture, Hyperspectral imaging, Quality assessment and grading.