Abstract:
Plant diseases cause significant agricultural losses worldwide, creating an urgent need for fast, accurate, and automated disease detection methods that can support effective crop management. Traditional diagnosis approaches rely heavily on expert knowledge and manual inspection, making them slow, subjective, and inefficient for large-scale farming. This research aims to address these limitations by developing an automated plant disease detection system using deep learning, specifically a Convolutional Neural Network (CNN), to classify visually similar plant diseases with improved precision and reliability.
The proposed CNN model incorporates multiple convolutional layers ranging from 32 to 512 filters, combined with max-pooling operations for spatial reduction and dropout layers to minimize overfitting. The final dense layer utilizes a SoftMax activation function to categorize images into their respective disease classes. Model training was conducted using the New Plant Dataset, consisting of over 80,000 RGB images across 38 disease and plant types. The Adam optimizer with a learning rate of 0.0001 and categorical cross-entropy loss function guided the training process. To enhance generalization, additional techniques such as data augmentation, batch normalization, and early stopping were employed.
Results demonstrated impressive performance, with the model achieving a 99% validation accuracy and consistently high classification capabilities across diverse plant disease categories. Evaluation metrics—including precision, recall, F1-score, and confusion matrix analysis—confirmed the system’s robustness and reliability. Overall, the developed system presents a practical and efficient framework for farmers, enabling early detection of plant diseases shortly after their onset. This supports timely intervention measures, improves crop protection, and contributes to sustainable agricultural productivity
