A MobileNet-enhanced diagnostic framework for accurate tomato leaf disease identification in smart farming

Plant diseases pose a major threat to global food security, causing up to 40% annual crop losses. Tomatoes are particularly vulnerable to fungal, bacterial, and viral infections that significantly reduce yield and quality. Deep learning has emerged as a powerful tool for automated plant disease diagnosis; however, many high-performing convolutional neural networks (CNNs) remain computationally expensive and unsuitable for mobile deployment. This work presents a comprehensive investigation of tomato leaf disease classification using a MobileNet-based transfer learning approach optimized for lightweight, real-time applications. The model was trained on the tomato subset of the PlantVillage dataset, comprising 18,345 images distributed across ten distinct classes. Extensive data augmentation, fine-tuning, and hyperparameter optimization were applied to enhance generalization. The proposed MobileNet model achieved a test accuracy of 95.76%, with macro-averaged precision, recall, and F1-scores exceeding 95%. ROC and Precision-Recall analyses yielded macro AUC values of 0.982 and 0.961, respectively, demonstrating strong discriminative capability. Comparative experiments showed that MobileNet provides the best trade-off between accuracy, parameter count, and inference speed when compared with MobileNetV2, MobileNetV3, ShuffleNet, and EfficientNet-B0. These results confirm that the proposed MobileNet model is a robust and efficient solution for tomato disease detection and is well-suited for real-world agricultural applications, particularly in resource-limited environments where mobile-based diagnostic tools can significantly support early intervention and reduce crop losses.