A Robust Ensemble Model for Plant Disease Detection Using Deep Learning Architectures

Abstract

This study explores advanced methods for plant disease classification by integrating pre-trained deep learning models and leveraging ensemble learning. After a comprehensive review of deep learning methods in this area, the InceptionResNetV2, MobileNetV2, and EfficientNetB3 architectures were identified as promising candidates, as they have been shown to achieve high accuracy and efficiency in various applications. The proposed approach strategically combines these architectures to leverage their unique strengths: the advanced feature extraction capabilities of InceptionResNetV2, the lightweight and efficient design of MobileNetV2, and the scalable, performance-optimized structure of EfficientNetB3. By integrating these models, the approach aims to improve classification accuracy and robustness and overcome the multiple challenges of plant disease detection. Comprehensive experiments were conducted on three datasets-PlantVillage, PlantDoc, and FieldPlant-representing a mix of laboratory and real-world conditions. Advanced data augmentation techniques were employed to improve model generalization, while a systematic ablation study validated the efficacy of key architectural choices. The ensemble model achieved state-of-the-art performance, with classification accuracies of 99.69% on PlantVillage, 60% on PlantDoc, and 83% on FieldPlant. These findings highlight the potential of ensemble learning and transfer learning in advancing plant disease detection, offering a robust solution for real-world agricultural applications.