Enhanced Rose Leaf Disease Classification Using Vision Transformer (ViT-B/16) Detecting Black Spot, Downy Mildew, and Healthy Leaves for Improved Plant Health Management

Though they are quite vulnerable to illnesses like Black Spot and Downy Mildew, rose leaves are essential for photosynthesis and general health of rose plants. Maintaining plant health and lowering economic losses in rose cultivation depends on early identification and classification of these diseases being effective. The Vision Transformer (ViT-B/16) model is investigated in this work for three classifications of rose leaves: Black Spot, Downy Mildew, and Fresh Leaf. Using transformer-based architecture, the ViT-B/16 model divides images into patches and captures global dependencies using self-attention methods. This method provides improved detection of minute changes in leaf patterns, therefore differentiating from traditional convolutional neural networks (CNNs.). From Kaggle, the dataset comprises 917 photos split into training, validation, and testing sets. The suggested approach calls for phases of data preparation, model construction, training, and evaluation. With high accuracy, recall, and F1-scores for Downy Mildew and Fresh Leaf categories as well as a somewhat lower but still strong performance for Black Spot, the ViT-B/16 model obtained an overall accuracy of 93%. Minimal overfitting and excellent learning and generalization are revealed by analysis of training and validation measures. The confusion matrix supports, even more, the great performance of the model in differentiating between disease types. The efficiency of the ViT-B/16 model in rose leaf disease classification is shown by this work, therefore stressing its possibilities for automated disease diagnosis and better agricultural practices.