In this study, we propose a two-stage deep learning framework for automatic detection and classification of diseases in lemon leaves. This system utilizes two convolutional neural networks based on MobileNetV2. The first model performs binary classification to distinguish between images of lemon leaves and irrelevant background images. The second model classifies the identified lemon leaf images into one of nine categories, including eight disease types and healthy leaves. The models are trained and evaluated using three publicly available datasets, with appropriate preprocessing and data augmentation. On the test dataset, the binary model achieves an accuracy of 99.7% with an F1-score of 0.99, while the multi-class classifier achieves 96% accuracy and an F1-score of 0.95. This approach was also compared with a single combined model of 10 labels, including a “Not-Lemon” category. That model also achieved a performance of 0.95 F1-Score, but took a longer time to converge. These results demonstrate the effectiveness of using a filtering stage with a multi-class classifier for plant disease detection tasks.
Muhammad Ishaque Junejo (Mon,) studied this question.