<![CDATA[Tomato cultivation is a vital agricultural commodity in Indonesia, yet leaf diseases continue to pose a serious threat to crop quality and yield. While deep learning–based classifiers have achieved high accuracy in laboratory settings, most existing tomato leaf disease detection models rely on computationally intensive architectures that limit their practical deployment on resource-constrained devices commonly used in agricultural environments. To address this gap, this study proposes a lightweight Convolutional Neural Network (CNN) based on the MobileNetV2 architecture, explicitly combined with systematic hyperparameter optimization, for tomato leaf disease classification. Using 14,529 images from the PlantVillage dataset, the research involves image preprocessing, data augmentation, and structured tuning to improve performance while maintaining computational efficiency. The optimized model achieves an accuracy of 81% using a learning rate of 0.001, 128 units, a dropout rate of 0.3, and an alpha value of 0.35. Although this accuracy is slightly lower than that reported by heavyweight CNN models, it is competitive for lightweight architectures and represents a favorable trade-off between classification performance and computational efficiency. Despite its compact design, the model demonstrates reliable disease recognition and suitability for deployment on devices with limited resources. Furthermore, the trained model was implemented in a desktop-based application as a proof-of-concept system, demonstrating scalability and potential adaptation to mobile or edge-based agricultural decision-support platforms. This study highlights the novelty of integrating lightweight CNN design with systematic hyperparameter optimization and demonstrates that optimized lightweight deep learning models can provide effective, efficient, and deployable solutions for real-world precision agriculture applications.]]>
