A HYBRID DEEP LEARNING FRAMEWORK FOR ACCURATE AND EFFICIENT DETECTION OF POTATO LEAF DISEASES

Despite significant advancements in deep learning-based plant disease detection, existing models often struggle with generalizability, computational efficiency, and adaptability to real-world agricultural challenges. To address these limitations, this study introduces a novel hybrid deep learning framework combining the custom DarkPotatoleafNet model with Vision Transformers (ViT) and convolutional neural networks (CNNs). Our approach leverages transfer learning and ensemble learning techniques to enhance disease classification accuracy while optimizing computational resources. The highest recorded accuracy of 99% was achieved through this hybrid architecture, demonstrating its superior performance over conventional models. To ensure optimal model training and generalization, we employed extensive preprocessing techniques, including data augmentation (rotation, scaling, flipping), grayscale conversion for structural emphasis, and image masking and segmentation to isolate diseased regions accurately. This research contributes to precision agriculture by providing a scalable and efficient AI-driven solution for early and accurate detection of potato leaf diseases, ultimately supporting sustainable farming practices and improved crop yield management.