Enhancing Rice Variety Identification through Fine-Tuning VGG-Based Deep Learning Models

Rice variety identification is critical for addressing global food challenges, given rice’s status as a staple food source for over half of the world’s population. Leveraging recent advances in machine learning, particularly deep learning models like VGG16 and VGG19, presents promising avenues for enhancing rice variety classification accuracy. This study delves into the refinement of rice variety identification through fine-tuning VGG-based deep learning models. By conducting a meticulous comparison of VGG16 and VGG19 models across transfer learning and fine-tuning phases, utilizing robust metrics such as accuracy, precision, recall, and F1-score, the research underscores the significant performance enhancements achievable through fine-tuning techniques. Notably, the VGG16 model exhibits exceptional accuracy, surpassing 99.96% after fine-tuning, showcasing a substantial improvement from the transfer learning phase. These findings highlight the efficacy of fine-tuning methodologies in optimizing classification performance and underline the transformative potential of deep learning models in agricultural classification tasks. The improvement observed from the transfer learning to the fine-tuning phase underscores the importance of iterative refinement in maximizing the models’ classification potential, indicating a promising direction for further research in agricultural classification methodologies.