Agriculture is one of society’s most important concerns, contributing approximately 80% of the world’s food supply. Unfortunately, plant diseases pose a serious problem and destroy crops every year. In particular, diseases affecting tomato crops lead to significant yield losses, impose greater economic burdens on farmers and countries, and damage the agricultural industry. This paper proposes an optimized model that uses various data augmentation techniques to preprocess datasets and enhance realism. It utilizes a convolution neural network (CNN) and multi-head attention model to classify diseases in tomato plants. Several experiments were conducted, and they provide evidence that applying preprocessing to the dataset using a simple CNN with attention mechanisms with optimized parameters has the potential to achieve high accuracy, with a 97.62% success rate of the test set in tomato classification problems. This proves the efficiency and feasibility of the proposed method.