In countries like India, agriculture plays a crucial role, but ensuring food security is a major challenge due to extensive crop losses caused by insufficient storage, transportation, and plant diseases. To deal with these issues, it is essential to create automated systems that can identify problems and assist farmers in implementing extensive strategies to reduce crop losses. There is a rise in employing deep learning models toward automatically detection of diseases in tomato leaves during last few decades. Ongoing research primarily centers on identifying different diseases, deficiencies, and factors affecting crop yield. This study explores various deep learning methods for detecting tomato diseases, utilizing a dataset consisting of eight tomato categories, with seven categorized as unhealthy and one as healthy. An extensive examination of the results underscores the superiority of transfer learning techniques over traditional deep learning methods. Especially, the InceptionV3 and VGG16 models show exceptional performance, achieving a notable accuracy of 94.7% in the detection of tomato leaves diseases and its classifications, surpassing prior benchmarks. We propose incorporating saliency maps as a visual tool to improve the interpretation of CNN classifications.