<![CDATA[Visual identification of various types of herbal leaves such as sweet leaf, moringa leaves, cat whiskers, bay leaf, and betel leaf is often difficult due to morphological similarities. This study presents a novel approach using a Convolutional Neural Network (CNN) model with the VGG16 architecture to automatically classify these five types of leaves. The main novelty of this study lies in the implementation of a two-stage fine-tuning strategy specifically tailored to the herbal leaf dataset. The first stage freezes the base layer and trains a new classification head, while the second stage fine-tunes several upper layers to improve model adaptability. The model was trained using 500 herbal leaf images and evaluated on 235 independent test images. The results demonstrated superior model performance with an overall accuracy rate of 91.06% and an average F1-score of 0.91. Qualitative analysis demonstrated the model's success in classifying leaves with unique features, such as cats whiskers and betel leaf. However, the model faced challenges in distinguishing leaves with high visual similarities, such as sweet leaf and moringa leaves. Practically, the developed model offers an effective and reliable solution for herbal leaf identification, reducing time and error rates compared to manual methods. Although this study is limited by the small dataset size, these results demonstrate the great potential of the optimized VGG16 architecture for applications in botany and traditional medicine, making it a valuable tool.]]>
