Smart plant disease diagnosis using multiple deep learning and web application integration

Abstract

Accurate and efficient plant disease diagnosis is crucial for sustainable agriculture and global food security, as diseases significantly impact crop productivity. Despite advancements in deep learning, the performance and scalability of many models remain limited. This study addresses this gap by evaluating MobileViTv2, EfficientNet-B7, and a hybrid MobileViTv2-EfficientNet-B7 approach for classifying plant leaf images into four categories: healthy, rust, scab, and multiple diseases. Using a publicly available dataset of annotated leaf images, the models were trained and tested under optimized conditions. MobileViTv2 emerged as the superior model, achieving the highest classification accuracy (94 %) and F1 score (0.94). It demonstrated exceptional generalization capabilities, with Receiver Operating Characteristic (ROC) Area Under Curve (AUC) values of 0.95 for healthy, 0.97 for rust, and 0.99 for scab. In contrast, EfficientNet-B7 and the hybrid model performed moderately, highlighting MobileViTv2's efficiency in handling diverse image features. To demonstrate real-world applicability, the MobileViTv2 model was deployed in a web-based application. This platform enables real-time plant disease diagnosis with high confidence, identifying conditions such as rust (85.3 % confidence) and healthy leaves (90.2 % confidence). The user-friendly interface facilitates its integration into precision agriculture. This study highlights the strengths of MobileViTv2 for disease diagnosis, its scalability, and its potential to support decision-making in agriculture. Future work will focus on expanding the model to other crops and incorporating environmental variables for enhanced disease prediction. This research bridges the gap between advanced AI models and practical agricultural applications, offering a robust solution for early disease detection. • Introduced MobileViTv2 for robust and efficient plant disease diagnosis. • Developed a MobileViTv2-EfficientNet-B7 hybrid model to address dataset imbalances. • Deployed a web-based application for real-time plant disease diagnostics. • Tackled dataset imbalance by advanced augmentation and hyperparameter optimization. • Demonstrated the lightweight and efficient design of MobileViTv2 for scalable use.

Topics & Concepts