Eggplant Disease Diagnosis Using a Robust Ensemble of Transfer Learning Architectures

Eggplant diseases present major challenges to agricultural productivity, leading to yield losses and threatening farmers livelihoods. Traditional diagnostic methods tend to be resource-intensive, subjective, and often inaccurate, making them unsuitable for large-scale use. This study addresses these shortcomings by proposing an Ensemble Transfer Learning (ETL) approach for precise classification of eggplant diseases. The model combines the strengths of ConvNeXt, EfficientNetV2, NFNet, and RegNet architectures, enhancing generalization and minimizing overfitting. Two datasets containing over 19,800 images were utilized for training and validation, with advanced preprocessing and data augmentation techniques applied to ensure high input data quality and model robustness. To assess performance, evaluation metrics such as accuracy, F1-score, MCC, and PR-AUC were used. The model achieved an impressive classification accuracy of 99.76%, surpassing state-of-the-art methods and demonstrating strong generalization across various conditions. A web-based application was also developed, enabling real-time disease diagnostics through a user-friendly interface. This application provides farmers with actionable insights, promoting sustainable farming practices by reducing crop losses and minimizing the need for chemical interventions. However, its reliance on high-quality labeled images limits its applicability in uncontrolled environments. Future research will aim to integrate more diverse datasets and develop mobile-based solutions to enhance scalability and usability. This study emphasizes the potential of ensemble transfer learning to transform agricultural disease diagnosis and improve crop productivity.