Deep Learning-Based Classification of Rice Varieties for Agricultural Applications

In the realm of precision agriculture, the classification of crop varieties holds significant importance. This study delves into the application of DL models for the classification of rice varieties, a critical aspect of agricultural research and practices. The research focuses on five rice varieties frequently cultivated in Turkey: Arborio, Basmati, Ipsala, Jasmine, and Karacadag.The study employs two distinct datasets. The primary dataset comprises 20,000 RGB images, with each variety contributing 5,000 images. Each image, encapsulating a single grain of rice, measures 250 × 250 pixels. The secondary dataset is a feature set, extracted from the grain images, consisting of 106 distinct features. These include 12 morphological, 4 shape, and 90 color features, providing a comprehensive and diverse data pool for the deep learning models.The DL models utilized in this research include ResNet50, MobileNetV1, VGG19, DenseNet, Inception v2, and AlexNet. The models exhibited high precision, recall, and F1-scores, with ResNet50 and DenseNet achieving an overall accuracy of 0.99. These results underscore the potential of deep learning in the accurate classification of rice varieties, thereby contributing to the enhancement of agricultural practices.However, the models displayed signs of overfitting, as evidenced by the lower validation accuracy compared to the training accuracy. To address this issue, the study explored several strategies, including the implementation of dropout layers and regularization.This research underscores the need for further studies to optimize these models and prevent overfitting. The findings pave the way for the development of more sophisticated and accurate models for rice variety classification, thereby contributing to the broader field of precision agriculture. This study stands as a testament to the potential of deep learning in revolutionizing agricultural practices and research.