Potato Disease Detection

Potato disease detection has emerged as a critical research domain due to the increasing need for precision agriculture and global food security. The rapid advancement of deep learning, particularly Convolutional Neural Networks (CNNs), Vision Transformers (ViTs), and multimodal architectures, has significantly improved the accuracy and scalability of plant disease detection systems. Recent studies propose hybrid transformer-based frameworks, ensemble CNNs, hyperspectral-CNN fusion, lightweight mobileoptimized models, IoT-integrated systems, and attentionenhanced architectures, achieving accuracies ranging from 96% to 99.4%. However, major limitations persist, including poor generalization to real-field images, computational complexity, dataset imbalance, reliance on controlled environments, and high hardware requirements. This review synthesizes findings from forty key research papers, highlighting their methodologies, datasets, strengths, and limitations. A detailed comparative analysis is presented to evaluate preprocessing methods, feature extraction strategies, real-field performance, and model robustness. The paper identifies critical research gaps such as the shortage of realfield datasets, lack of domain adaptation, challenges in mobile deployment, and limited multimodal integration. Based on these gaps, a conceptual methodology is proposed to enhance real-field disease prediction through hybrid architectures, self-supervised learning, and adaptive preprocessing. Finally, the review outlines future research directions focused on lightweight transformer models, sensor fusion, cross-farm federated learning, and explainable AI for trustworthy agricultural applications. This work aims to guide researchers and developers in designing the next generation of scalable, efficient, and farmer-centric potato disease detection systems.