Optimizing Rice Variety Recognition Through Transfer Learning with Pretrained ResNet Models

The identification of rice varieties holds paramount importance in addressing global food challenges, given that rice serves as a staple food source for more than half of the world’s population. Accurate identification is crucial for ensuring food security and meeting the dietary needs of diverse populations. However, accurately distinguishing between rice varieties remains a complex task due to genetic mixing and hybridization, which can lead to variations in morphological and genetic traits among different varieties. Recent advancements in deep learning, particularly convolutional neural networks (CNNs), offer promising solutions for improving rice variety classification by leveraging large datasets and sophisticated image processing techniques. This study conducts a comprehensive comparative analysis of ResNet50 and ResNet101 models in identifying rice varieties using transfer learning and fine-tuning approaches. Leveraging a dataset containing 75,000 high-resolution rice images representing various varieties, the models undergo thorough evaluation based on various performance metrics, including accuracy, precision, recall, and F1-score. Results highlight the effectiveness of pretrained ResNet models in accurately classifying rice varieties, with ResNet50 achieving the highest accuracy of 98.813% post-fine-tuning. These findings underscore the significance of fine-tuning pretrained models to enhance classification accuracy, contributing to the advancement of precision agriculture and sustainable food production practices. The insights gained from this study have implications for improving crop management practices, optimizing resource allocation, and ultimately, enhancing global food security.