A Transformer-based Ensemble Model for Plant Disease Detection

Plant diseases significantly affect productivity on agriculture, leading to financial losses and food safety issues. This study analysed the efficiency of models trained with deep learning using the PlantVillage dataset for plant disease classification and detection. Various models and ensemble learning techniques, including vision transformers, swin transformers, and MobileNetV2, have been analysed for their efficiency in feature extraction and classification. It was observed that the ensemble model, swin transformer, and mobilenetv2 yielded the highest accuracy of 98.92%, outperforming individual models such as swintransformers (96.83%), vision transformers (96.55%), mobilenetv2(94.15%), and the ensemble model of the swintransformer and xgboost(95.99%). These findings establish a foundation for AI-driven precision farming tools that enable early disease detection and minimise crop loss.