Light-Weight Deep Convolutional Neural Network Model for Classification of Potato Leaf Diseases

Potato leaf diseases pose a significant threat to global food security by reducing crop yields and economic productivity. Traditional manual inspection methods are often inefficient and error-prone, particularly in developing countries. Automated deep learning approaches provide a promising alternative for accurate and timely disease detection. This study develops a lightweight deep convolutional neural network (DCNN) for classifying potato leaf diseases, including early blight, late blight, and healthy leaves, while ensuring high accuracy, efficiency, and deployability on edge devices. A dataset of 2,152 potato leaf images, sourced from Kaggle, was preprocessed, augmented, and partitioned into 80% training, 10% validation, and 10% testing sets. A custom DCNN architecture (2.2M trainable parameters) was designed and compared against Xception, ResNet50, and InceptionV3 using precision, recall, F1-score, specificity, accuracy, and Cohen’s Kappa metrics. The proposed model outperformed existing architectures, achieving 97.21% accuracy, 93.92% F1-score, 95.83% precision, 92.33% recall, 98.38% specificity, and 95.00% Kappa score, with a compact size of 25.6 MB. Deployment on a Streamlit-based web application demonstrated real-time classification capabilities, achieving near-perfect accuracy (99.99%) for early and late blight detection. The lightweight DCNN offers an efficient, accurate, and deployable solution for potato disease classification, suitable for edge devices such as smartphones. This system empowers farmers with rapid, automated diagnostics, enabling timely interventions to mitigate crop losses. Future work will focus on extending the model to additional potato species and optimizing deployment for mobile platforms.