Deep Learning-Based Classification of Rice Varieties: A Comprehensive Study

Abstract

The classification of rice varieties is a critical task in agriculture, playing a pivotal role in quality control, yield optimization, and trade. Traditional methods for rice classification often rely on manual inspection, which is time-consuming and prone to human error. This paper investigates four important convolutional neural networks (CNNs) for rice classification. In this study, five distinct rice varieties, namely Basmati, Jasmine, Arborio, Karacadag, and Ipsala, were examined. A dataset comprising 75,000 rice images, with 15,000 images per rice type and a uniform image size of 250 × 250 pixels, is utilized for the investigation. To build classification models, VGG-16, VGG-19, ResNet50, and InceptionV3 architectures are employed. These models are subsequently applied to classify rice images, and their performance is evaluated using various statistical metrics, including accuracy, confusion matrix, recall, precision, and F1-score. These results underscore the effectiveness of these deep learning (DL) models in accurately categorizing different rice varieties. The findings of this study demonstrate the practical utility of these models for automating rice variety classification, which can have significant implications for quality control and classification processes in the rice industry.