KLASIFIKASI DAUN HERBAL MENGGUNAKAN METODE CONVOLUTIONAL NEURAL NETWORK DENGAN ARSITEKTUR VGG-19

Indonesia has a wealth of biodiversity, including a variety of medicinal plants that play a significant role in herbal medicine. The increasing interest in the use of herbal medicine drives the need for an automatic classification system capable of accurately recognizing plants. This research proposes a method using Convolutional Neural Network (CNN) with the VGG19 architecture to classify herbal leaves based on images. The applied modifications include the implementation of Adam optimization, early stopping techniques to prevent overfitting, and the application of color space conversion from RGB to HSV to enhance the model's performance. The research process includes inputting the dataset, pre-processing, data splitting, model implementation, and model evaluation and testing. The dataset used is the "Indonesian Herb Leaf Dataset 3500" from the Mendelay Data site, consisting of 3500 herbal leaf images and containing 10 classes: starfruit leaf, guava leaf, lime leaf, basil leaf, aloe vera, jackfruit leaf, pandan leaf, papaya leaf, celery leaf, and betel leaf, with each class containing 350 images. Testing results show that the VGG19 architecture with the application of color saturation conversion from RGB to HSV achieved a training accuracy of 97.36% and a testing accuracy of 99%. The test results indicate that the performance of the VGG19 model is better than that of VGG16 in the herbal leaf classification study.