Most of the diseases in tomatoes can be determined by observing the leaves. Therefore, the observation of tomato leaves is of critical importance in vegetable cultivation. Traces on the leaves, color changes, curling, etc. may indicate a disease or a virus that may pose a risk of disease. Since preserving the health of the leaves will increase the productivity of tomatoes, it is a necessity to observe the leaf development and to determine the possible disease risks accurately and quickly in advance. In the proposed study, the "T-LeafNet" Convolutional Neural Network (ESA-CNN) model, which can detect 9 different diseases and healthy leaves in the leaves of tomato plants, is proposed. Classification success was measured by training the proposed network from scratch. In addition, within the scope of transfer learning, MobileNetV2, AlexNet and VGG16 architectures, which are ready-made network models, were also tested on the dataset and the results were compared. The effects of data augmentation and learning coefficient parameters on the achievement were also examined by applying the tests on the original and augmented data with different learning coefficients. The highest classification performances of T-LeafNet, AlexNet, MobileNetV2 and VGG16 networks were measured as 97.32%, 98.32%, 99.1% and 99.21%. Considering the training time and classification performances, it has been observed that the proposed T-LeafNet network gives high performance with shorter time and less resources compared to other tested networks.