Research on Plant Disease Recognition Using Few‐Shot Learning

Plant diseases pose a significant threat to global agriculture, leading to substantial crop losses and economic damage. Traditional deep learning methods for plant disease recognition require large labeled datasets, which are often unavailable for rare or emerging diseases. This thesis addresses the challenge of data scarcity by proposing a few‐shot learning approach using Siamese Networks for plant disease recognition. The study leverages the PlantVillage dataset, applying advanced preprocessing and data augmentation techniques to enhance model robustness. The Siamese Network architecture is designed with twin convolutional networks sharing weights, trained using contrastive loss to measure similarity between image pairs. Experimental results demonstrate the modelʹs effectiveness in classifying plant diseases with limited labeled examples, achieving competitive accuracy compared to traditional CNN‐based methods. The framework is further evaluated through ablation studies, highlighting the impact of data augmentation, pair selection strategies, and hyperparameter tuning. Additionally, a prototype visualization system is developed to provide interpretable results for real‐world agricultural applications. The system’s deployment potential is explored in precision agriculture, mobile applications for smallholder farmers, and large‐ scale disease surveillance networks. The research contributes to sustainable farming practices by enabling early and accurate disease detection with minimal data, offering a scalable solution for resource‐constrained environments.