Improving Potato Disease Classification with Convolutional Neural Networks and Feature Fusion

Accurate prediction of potato diseases plays a vital role in maintaining crop health and maximizing agricultural yield. This study presents a deep learning approach utilizing Convolutional Neural Networks (CNNs) integrated with advanced feature fusion strategies to enhance the precision of disease classification. The methodology employs three feature fusion strategies—Early, Intermediate, and Late Fusion—on a benchmark dataset of 2,152 potato leaf images categorized into Early Blight, Late Blight, and Healthy. The workflow includes data preprocessing, feature extraction using CNNs, and classification through fully connected layers.Results indicate that Late Fusion outperforms other strategies, achieving the highest classification accuracy and robust generalization despite class imbalance. It effectively distinguishes between Early and Late Blight, while Early Fusion shows better performance for Healthy leaves. The study highlights the importance of feature fusion and data augmentation in enhancing disease diagnosis, paving the way for scalable and efficient agricultural solutions.