Plant diseases continue to be a chronic threat to world food security as they affect crop yields and quality. This research seeks to solve the outstanding issue of delayed and imprecise diagnosis of plant diseases by proposing an AI-based, image-based detection framework. The system combines deep learning architectures, sophisticated image processing, and a mobile interface to detect plant diseases, quantify their severity, and suggest tailored interventions. Utilizing EfficientNetB0 trained on a multi-crop dataset of high-resolution leaf images, the model attains high disease classification accuracy for apple, corn, and grape diseases. Severity indices of disease quantify the extent of infection, allowing accurate pesticide application and sustainable resource management. The study employs an experimental setup with an 80–20 train-validation split, data augmentation methods, and severity-weighted loss optimization. Empirical findings show model accuracies ranging from 94% to 98% for apples and grapes, and 87% for maize, underpinned by confusion matrices and F1-scores. This research makes contributions both theoretically—by improving disease classification through severity weighting—and managerially—by providing farmers with real-time, mobile-enabled diagnostic capabilities. Implications arise for precision agriculture, sustainable agriculture, and AI-driven agricultural policy planning.