The advancement of deep learning has significantly improved the automation of plant disease detection through image classification. This study compares the performance of standard DenseNet121 and an enhanced version incorporating Squeeze-and-Excitation (SE) blocks for classifying tomato leaf diseases. A dataset derived from PlantVillage was used, covering multiple disease categories and healthy leaves. To improve generalization, extensive data augmentation techniques were applied. Both architectures were implemented and trained using PyTorch, with evaluation metrics including accuracy, precision, recall, F1-score, and inference time. The experimental results demonstrate that DenseNet121-SE significantly outperforms the standard DenseNet121, achieving a classification accuracy of 99.00%. The integration of SE blocks allows the model to recalibrate channel-wise features adaptively, enhancing sensitivity to important patterns while maintaining computational efficiency. This study highlights the effectiveness of attention mechanisms and data augmentation in improving classification performance and supports their practical application in intelligent agriculture systems.