Classification of Balinese Herbal Plant Leaf Using Convolutional Neural Network

Abstract

This research initiates the classification of Balinese herbal leaves based on Usada Taru Pramana Manuscript, one part of Balinese local cultural wisdom that contains knowledge about herbal plants and is used traditionally for treatment. The challenge in this research is that there is no study on herbal leaf image dataset based on Balinese herbal leaves available for research. Therefore, in this research we collected 50 types of Balinese herbal plants which taken from various regions in Bali. The classification method used relies on the Inception ResNet V2 CNN architecture as well as data augmentation techniques to increase the limited amount of training data. The training process uses 3000 images, while validation uses 200 images, and the testing process is performed on 200 images. The experimental results prove that this method produces good accuracy in the classification of herbal leaf types, which is 99.73% for training data, 67.71% for validation data, and 72.50% for testing data. The results of this research can be used to support the development of automation systems that help humans recognize herbs more precisely and quickly, as well as facilitate the preservation and utilization of cultural heritage in traditional medicine and botanical science.