This study presents a deep learning-based approach for accurate diagnosis of sugarcane leaf diseases using Vision Transformers (ViT) and Convolutional Neural Networks (CNN). The research evaluates the performance of various models, including DeiT3-Small and DeiT-Tiny ViT architectures, and CNN models such as ResNet50, Xception, and EfficientNetv2-Small, on the publicly available Sugarcane Leaf Dataset, which contains 6748 images of sugarcane leaves across 11 disease classes. The results show that the DeiT3-Small model achieved the highest accuracy (93.79%), precision (91.27%), and F1-score (90.96), outperforming other models. The study highlights that deep learning techniques can significantly improve sugarcane disease management and contribute to increased yields. The findings also indicate that there is no direct relationship between model complexity and accuracy for the 11-class sugarcane dataset. The research demonstrates the effectiveness of data-efficient ViT models in image classification tasks, showing that they can achieve high performance with fewer parameters. The study also discusses the importance of image processing techniques in plant disease detection and the potential of deep learning in improving agricultural productivity and sustainability. The results suggest that both CNN-based and ViT-based models can be used effectively in various visual tasks, with ViT-based models showing high F1-scores and competitive accuracy. The study concludes that the choice of model depends on the specific requirements of the task, and further research is needed to validate the proposed models in real-world applications.This study presents a deep learning-based approach for accurate diagnosis of sugarcane leaf diseases using Vision Transformers (ViT) and Convolutional Neural Networks (CNN). The research evaluates the performance of various models, including DeiT3-Small and DeiT-Tiny ViT architectures, and CNN models such as ResNet50, Xception, and EfficientNetv2-Small, on the publicly available Sugarcane Leaf Dataset, which contains 6748 images of sugarcane leaves across 11 disease classes. The results show that the DeiT3-Small model achieved the highest accuracy (93.79%), precision (91.27%), and F1-score (90.96), outperforming other models. The study highlights that deep learning techniques can significantly improve sugarcane disease management and contribute to increased yields. The findings also indicate that there is no direct relationship between model complexity and accuracy for the 11-class sugarcane dataset. The research demonstrates the effectiveness of data-efficient ViT models in image classification tasks, showing that they can achieve high performance with fewer parameters. The study also discusses the importance of image processing techniques in plant disease detection and the potential of deep learning in improving agricultural productivity and sustainability. The results suggest that both CNN-based and ViT-based models can be used effectively in various visual tasks, with ViT-based models showing high F1-scores and competitive accuracy. The study concludes that the choice of model depends on the specific requirements of the task, and further research is needed to validate the proposed models in real-world applications.