<![CDATA[For conservation, commercial cultivation, and scientific research, accurate identification of orchid species often requires specialized expertise. In this study, the DenseNet121 deep learning architecture was employed to develop an automated classification system for four popular orchid species. DenseNet121 was selected for its ability to extract complex hierarchical features and its strong performance on limited-scale datasets. The initial dataset comprised 1,935 images of Phalaenopsis, Cattleya, Dendrobium, and Vanda orchids. However, after manual removal of duplicate images, only 1,658 images remained, revealing significant class imbalance. The undersampling method was applied to balance each class to 248 samples. The dataset was then split into 75% training, 15% validation, and 10% testing, and enhanced through data augmentation techniques such as rotation, flipping, brightness variation, width shift, height shift, and zoom. The final model achieved 97.00% accuracy with class-specific performance ranging from 92.59% to 100% accuracy across different orchid species. This research can serve as a foundation for developing mobile or web applications to assist researchers, farmers, and orchid enthusiasts in accurately identifying orchid species, while supporting conservation efforts for orchid biodiversity in Indonesia.]]>
Copyrights © 2025
