<![CDATA[Herbal plants represent one of Indonesia's rich biodiversity resources that have long been utilized in traditional medicine. However, manual identification remains challenging due to morphological similarities among plant species. Various studies have applied Convolutional Neural Network (CNN) for herbal plant classification, yet comparative analysis between Inception-v3 and Inception-v4 in this domain remains limited. This comparison is necessary as increased architectural complexity in Inception-v4 does not always guarantee better performance on small-scale datasets. This study aims to compare the performance of Inception-v3 and Inception-v4 transfer learning in classifying 10 herbal plant species using 1,000 leaf images. The novelty lies in a direct comparative analysis considering data augmentation and hyperparameter tuning. Pre-processing includes image resizing and augmentation, while hyperparameter tuning applies learning rate variations (0.001; 0.0001; 0.00001) and batch sizes (16, 32, 64). Evaluation was conducted using accuracy, precision, recall, and F1-score. Inception-v3 achieved the best performance with 98.50% accuracy, 98.55% precision, 98.50% recall, and 98.50% F1-score, providing an empirical benchmark for Inception architecture selection in leaf-based herbal plant classification.]]>
