Deep Learning-Based Rice Classification Using VGG-16: A High-Accuracy Approach with Kaggle Dataset

The research of classification of rice is highly essential in assessing agricultural quality and identifying varieties. This paper attempts to analyze the potential of deep learning (DL) models, such as VGG-16, in classifying rice varieties from image-based datasets. The dataset was taken from Kaggle with five varieties: Arborio, Basmati, Ipsala, Jasmine, and Karacadag, including 15,000 high-resolution images. Techniques used for preprocessing the data were resizing, normalization, and augmentation to increase model robustness. This proposed model uses VGG-16 with transfer learning and is optimized using the Adam optimizer at a learning rate of 0.001 over ten epochs. High classification accuracy results are achieved by the model at 99.59%, where it surpasses traditional machine learning (ML) techniques like SVM and Random Forest. The confusion matrix also has minimal misclassifications, showing precision, recall, and F1-score at 1.00 for all classes, hence the model. The lightweight models still pose challenges in diversity of datasets and computational efficiency despite their superior performance. Future work should be to optimize the models and expand datasets for broader application in real world agricultural settings. This study gives an insight into the potential use of DL for automated classification of rice and contributes to precision agriculture and efficient grain quality assessment.