Coffee rust is a coffee disease caused by the fungus Hemileia vastatrix. Threatening crops globally and causing economic losses to coffee growers. Early and accurate detection of this disease is critical for its effective control. The objective of this study was to classify coffee leaves with rust automatically by applying deep learning techniques. The methodology employed includes four (04) phases: obtaining the dataset; preprocessing (Normalization, Segmentation and Scaling); modeling and implementation (Vision Transformer, NASNet, VGG19 and ResNet50), and evaluation (Accuracy, Precision, Recall and F1 Score). Superior results were obtained with the Vision Transformer: accuracy of 92.90%, recall of 92.04%, F1 Score of 91.75%, and ROC of 0.99, indicating a high capacity to discriminate between healthy and diseased leaves. These results highlight the potential of the Vision Transformer as a valuable technique for the early detection of coffee rust, allowing coffee farmers to take timely control measures, reduce losses, and optimize crop productivity.