A Hybrid Deep Learning Model for Potato Leaf Disease Classification using Transfer Learning and Attention Mechanism

Potato leaf diseases pose a significant threat to crop yield and quality, necessitating accurate and timely classification for effective disease management. Deep learning has emerged as a powerful tool for automated disease detection, enabling precise classification of infected leaves. This work proposes a novel hybrid deep learning framework integrating DenseNet- 169, a Group Shuffle Depthwise convolution (GSDW) block, and Triplet Attention blocks to enhance feature extraction for better classification accuracy. DenseNet-169 serves as the backbone for deep feature extraction, using its dense connectivity for improved gradient flow and feature reuse. The GSDW captures cross- channel dependencies, modeling complex relationships between disease features. Additionally, the Triplet Attention module re- fines feature selection by focusing on relevant disease patterns, ensuring accurate classification. By combining global feature aggregation with localized attention mechanisms, the proposed model achieves better performance in distinguishing various potato leaf diseases. The proposed model attains an accuracy of 91.7%, outperforming existing methods and offering a reliable solution for automated potato leaf disease diagnosis.