This study proposes an integrated approach, employing image processing along with neural network learning methods for the classification of disease in jasmine leaf. The augmented dataset comprises 3537 images categorized into three classes Healthy, Resistant, and Susceptible. A Neural Network (CNN) is employed to detect and classify diseases, analyzing a comprehensive dataset spanning from initial healthy stages to advanced infection stages. Healthy Jasminum Sambac is characterized by vibrant green leaves, vigorous growth, and fragrant flowers. As Alternaria leaf blight disease progresses, it becomes more noticeable and widespread, leading to severe damage in advanced stages. Early detection and timely actions are essential for safeguarding the plant's health. CNN and VGG16 (visual geometry group) techniques are assessed through different deep learning methods, including custom learning and pre-trained transfer learning using (CNNs) architecture. The results demonstrate effective image processing to optimize the elimination of background noise before training a suitable deep learning (DL) model. The deep learning model with pre-trained transfer learning achieves an 87.83% accuracy for 20 epochs, while the custom CNN model achieves a 95.57% accuracy for 100 epochs, optimally classifying defective leaves into three stages (Healthy, Beginning, and Final) with an overall 0.9557 Mean Squared Error (MSE). This ensures that the CNN-based model can effectively distinguish fungal diseases in leaves. These results showcase the model's robustness and accuracy in predicting diseases. The efficiency of the developed CNN model is further assessed by confusion matrix.