<![CDATA[Indonesia is renowned for its rich diversity of spices, which hold significant cultural and economic value. However, public knowledge of these spices remains limited, making their identification challenging. Addressing this issue, this study aims to develop a scalable spice identification system using Convolutional Neural Networks (CNN) with a Transfer Learning approach. The system is designed to recognize 30 types of spices while maintaining high accuracy, utilizing the Cross-Industry Standard Process for Data Mining (CRISP-DM) framework for systematic development. The dataset was collected through open sources and web scraping from Google Images. Four CNN models (ResNet50, EfficientNetB0, Xception, and MobileNet) were evaluated under three data splits: 90:10, 80:20, and 70:30. Performance metrics including accuracy, precision, recall, and F1-score were used for evaluation. Among these models, Xception achieved the best performance in the 90:10 split, with an accuracy of 84.51%, followed by EfficientNetB0 at 83.57%. The results demonstrate that transfer learning effectively enhances model accuracy and scalability, enabling reliable spice identification across diverse categories. This system has practical implications for promoting public awareness, supporting culinary industries, and preserving Indonesia’s rich spice heritage. The proposed approach highlights the potential of CNN-based systems for addressing classification challenges in resource-constrained settings, offering a foundation for future research and real-world applications.]]>
Copyrights © 2025
