Klasifikasi Penyakit Tanaman Daun Tomat Menggunakan Transfer Learning

Abstract

Leaf diseases in tomato plants can significantly reduce crop yields and threaten agricultural sustainability. This study proposes a multi-class classification approach for tomato leaf diseases using transfer learning with pre-trained CNN architectures, specifically DenseNet121 and DenseNet169. The dataset used is a subset of PlantVillage, consisting of six disease classes and healthy leaves, with preprocessing steps including image augmentation and resizing. The training strategy involves two phases: feature extraction and fine-tuning, optimized using the Adam algorithm and categorical cross-entropy loss function. Evaluation metrics such as accuracy, precision, recall, and F1-score show that the DenseNet121 model achieves the best performance, reaching an accuracy of 96.23%, followed by MobileNetV2 with 92.89%. Loss curves and confusion matrix analysis confirm that the model performs classification tasks with stability and high precision, despite some misclassifications between visually similar disease classes. This study demonstrates that transfer learning with DenseNet—particularly DenseNet121 is effective for automatic and efficient classification of various tomato leaf diseases, offering potential for real-world implementation as a computer vision-based plant disease diagnosis system.