CNN-Based Eucalyptus Disease Classification Using MobileNet

Eucalyptus pellita is a key species in Indonesia’s pulp and paper industry. However, it remains vulnerable to various foliar diseases, including Xanthomonas infections and pest infestations, particularly those caused by leaf roller caterpillars, which significantly reduce nursery productivity. This study presents a leaf disease classification system utilizing a Convolutional Neural Network (CNN) based on the MobileNet architecture and a transfer learning approach. A dataset of 600 images was compiled and categorized into three classes: healthy leaves, Xanthomonas-infected leaves, and caterpillar-damaged leaves. The dataset was divided into training (80%), testing (10%), and validation (10%) subsets. Multiple hyperparameter configurations were tested, with the optimal performance achieved using a learning rate of 0.0001, 20 training epochs, a batch size of 32, and the Adam optimizer. The resulting model achieved 97.29% training accuracy and 100% accuracy on both the validation and testing sets, demonstrating strong reliability in detecting leaf diseases. These findings provide a solid foundation for the development of technology-assisted disease management tools to support agronomic decision-making and agricultural extension services.