Application of Deep Learning for Mobile-Based Herbal Leaf Diseases Classification Using CNN-Based Ensemble Learning

Herbal leaves are an integral part of traditional Balinese medicine, known for their natural healing properties with minimal side effects. However, limited knowledge of the diseases that attack these plants may lead to decreased use and conservation. This research aims to develop a disease classification model on herbal plants using Convolutional Neural Network (CNN) as well as ensemble learning methods. The CNN method used to automatically extract features from leaf images, while ensemble learning using the ensemble averaging method improves the accuracy of classification results. This research identifies various leaf diseases, such as leaf spot, bacterial spot, mosaic virus, powdery mildew, and target spot, as significant threats to these crops. The proposed model was implemented in a mobile application, allowing users to detect diseases directly through their devices, thus supporting the preservation of the Usada Bali treatment tradition. The ensemble model, which incorporates architectures such as ResNet50, DenseNet121, MobileNetV2, and VGG16, achieved high accuracy in validation, with results up to $96.7 \%$. The findings show that this model offers a practical solution for the early classification of diseases in herbal plants, contributing to the conservation of traditional Balinese medicine.