Leveraging Pre-trained Deep Learning Models for Orange Leaf Disease Classification

The research work investigates the use of pre-trained deep learning models, VGG16 and EfficientNetB3, for multi-class classification of orange leaf diseases. The paper aims at improving the accuracy and efficiency of disease diagnosis in orange crops which is also a contribution to the development of agricultural diagnostic tools. Using intensive data preprocessing, model adaptation, and fine-tuning, the research shows the models' ability to detect five specific disease classes as well as a healthy leaf. The VGG16 model achieved an overall accuracy of 89.5%, whereas the EfficientNetB3 model exceeded this with an accuracy of 92.3%, which demonstrates that the latter is better at capturing disease patterns with an optimally balanced network. The findings demonstrate that pre-trained models can be used in plant pathology and provide avenues for new research in agricultural technology. Through the strengths of VGG16 and EfficientNetB3, this study underlines the great advantages of deep learning models in disease detection procedures to promote sustainable and productive agricultural practices.