Optimizing rice variety classification with transfer learning: A robust and scalable model for agricultural applications

This paper investigates the use of transfer learning to classify five rice varieties using images of rice grains and husks. Automated rice variety classification is a crucial component of precision agriculture, offering numerous benefits to optimize rice production. Accurate classification enables farmers to make informed decisions about seed selection, crop management, and quality control. Automated systems, powered by computer vision and deep learning, offer a more efficient and reliable solution compared to traditional methods, enabling rapid processing of large datasets, improved accuracy, and real-time monitoring. The authors compare six pre-trained deep convolutional neural networks (VGG19, Xception, ResNet50, ResNet152V2, DenseNet201, and EfficientNetB7) and propose a fine-tuned ResNet50-based model. The experimental results demonstrate that the proposed model achieves the highest accuracy (98.12%), outperforming the other transfer learning models considered and improving the previous work in the literature. This approach offers a promising solution for automated rice variety classification, contributing to precision agriculture. • Uses husk image analysis with fine-tuned ResNet50 and transfer learning for accurate, scalable rice variety classification. • Husk images provide key structural and textural features for robust rice variety classification in real-world conditions. • Fine-tuned ResNet50 with custom layers, dropout, batch norm, and L2 reduces overfitting, achieving 98.12% accuracy. • Transfer learning on pre-trained DCNNs enables accurate, scalable rice variety classification even with limited data.