The process began with selecting a high-quality, labeled dataset (PlantVillage) that included images of healthy and infected leaves across multiple crop types. We cleaned and preprocessed the dataset by resizing, normalizing, and augmenting the images using rotation, zoom, and flipping to increase data diversity. Next, we designed a Convolutional Neural Network with stacked convolutional and pooling layers, followed by dense layers and a softmax output for multiclass classification. The model was trained using categorical cross-entropy and optimized with the Adam optimizer.

We evaluated the model using validation data and fine-tuned it by experimenting with batch size, dropout, and learning rate. The system was able to classify leaf diseases with a high level of accuracy and minimal overfitting. We also analyzed misclassified cases to further optimize performance. Finally, the model was packaged into a simple interface where users could upload a leaf image and instantly receive disease predictions.