<p>Tomato is one of the most widely consumable vegetable worldwide. Since, their yield and quality are significantly affected by the atmospheric conditions, cultivation practices, fertilization, watering, diseases, and harvesting methods. A scalable diagnosis of tomato leaf diseases is essential for sustainable agriculture. This research investigates the role of model optimization in improving classification performance by evaluating two convolutional neural networks (CNNs) frameworks: (i) a baseline CNN trained with the Adam optimizer, and (ii) an enhanced CNN trained with Sharpness-Aware Minimization (SAM) (Foret et al. in Sharpness-aware minimization for efficiently improving generalization, 2020). The work propose the use of Sharpness-Aware Minimization (SAM) for the better generalization by optimizing flatter minima in the loss landscape and this address the limitations of traditional optimizers like Adam that converge to sharp minima prone to overfitting. The min–max (Foret et al. 2020) enhances the model robustness on heterogeneous agricultural datasets. The models were trained and tested on a dataset of 157,871 tomato leaf images across 12 classes, The models have shown reliable performance, but the SAM based model outperformed the baseline models in terms of generalization and faster convergence. The results provide methodological insights of the optimizer impacts on plant disease classification and underline applicability in precision agriculture and prove that the integration of SAM with CNN training improves the generalization, robustness, and convergence speed.</p>