Deciphering Indian Peppermint Classes: CNN and Decision Trees Fusion

The proposed research explores a robust methodology for the classification of peppermint varieties in India, employing Convolutional Neural Networks (CNNs) in conjunction with Decision Trees (DT). A dataset comprising approximately 13,000 high-resolution peppermint images, partitioned into training, validation, and testing subsets, facilitated model development and evaluation. The proposed framework achieved an impressive overall accuracy rate of $\mathbf{9 8. 2 \%}$ in distinguishing between Kukrail, CIM-Patra, and CIM-Indus peppermint varieties. Performance parameter analyses, detailed in a tabular format, underscored the models’ consistent precision, recall, and F1 scores across the diverse peppermint classes. However, the Confusion Matrix revealed occasional misclassifications, especially between CIM-Patra and CIM-Indus, warranting further investigation into finer differentiators. This study highlights the potential of CNNs and DT in agricultural classification tasks, specifically in peppermint variety identification. The fusion of these methodologies demonstrates promising results, enabling precise variety categorization crucial for tailored agricultural practices, enhanced crop management, and improved product quality. The findings provide a foundation for leveraging advanced machine learning approaches in agricultural contexts, with implications extending to optimized farming practices and informed decision-making in peppermint cultivation and potentially other agricultural sectors. Future directions involve refining models through feature engineering, ensemble methods, and broader dataset integration to enhance robustness and generalization.