Tomato plants are crucial but prone to diseases that significantly impact yield and quality. Early detection is vital to mitigate agricultural losses. In this study, we propose deep learning approaches to classify tomato leaf diseases using ResNet50 and VGG16 convolutional neural networks (CNNs), focusing on architectural optimization and fine-tuning for enhanced performance. For ResNet50, custom layers, including global average pooling, dropout, and dense layers, achieved 96.62% accuracy with frozen pre-trained layers. Fine-tuning the Conv5 block and adjusting hyper parameters improved validation accuracy to 98.78%, with notable gains in precision and recall. The VGG16 model, augmented with dense and dropout layers, initially achieved 96.34%. Fine-tuning blocks 4 and 5 and incremental learning rate reductions improved accuracy to 99.08%, demonstrating superior generalization. This research underscores the efficacy of optimized CNN architectures and fine-tuning techniques in achieving high-precision automated detection of tomato leaf diseases.