Deep Learning-Based Classification of Potato Leaf Diseases Using a Custom Convolutional Neural Network

Early and accurate identification of potato leaf diseases is critical for minimizing crop losses and ensuring sustainable agricultural productivity. While deep learning–based approaches have demonstrated high classification accuracy, many existing solutions rely on deep and computationally expensive architectures that are unsuitable for deployment in resource-constrained agricultural environments. This study presents an efficient and lightweight Convolutional Neural Network (CNN) designed for real-time classification of potato leaf diseases, specifically Early Blight, Late Blight, and Healthy leaf conditions. The proposed architecture integrates Batch Normalization to stabilize learning dynamics and accelerate convergence, along with Global Average Pooling to significantly reduce parameter complexity. Experimental evaluation demonstrates that the model achieves a validation accuracy of 97.62% within only 10 training epochs, while maintaining high precision, recall, and F1-score across disease classes. The results confirm that carefully designed shallow CNN architectures can deliver performance comparable to deeper networks with substantially lower computational overhead, making the proposed approach well-suited for edge-based and real-time agricultural applications.