Journal of Agronomy and Field Innovations

Background: Rice (Oryza sativa L.) is the staple food crop for over 3.5 billion people globally, yet rice leaf blast caused by the fungal pathogen Mangalore Oryza is a devastating disease-causing annual yield loss exceeding 30% in susceptible cultivars. Conventional disease diagnosis relies on visual assessment by trained personnel, which is time-consuming, subjective, and limits early intervention opportunities.
Objective: To develop and evaluate a deep learning-based convolutional neural network (CNN) model for automated, rapid detection and classification of rice leaf blast at early disease stages, enabling timely disease management intervention.
Methods: A comprehensive dataset of 8,500 rice leaf images (healthy, n=2,800; early blast, n=2,850; moderate blast, n=1,600; severe blast, n=1,250) was collected from field and controlled environments at 300 dpi resolution. Images were preprocessed through resizing, normalization, and augmentation. A ResNet50 architecture was trained with transfer learning using ImageNet pre-trained weights, batch size 32, initial learning rate 0.001 (Adam optimizer), for 150 epochs on a GPU-accelerated computing platform. Data were partitioned into 70% training (5,950 images), 15% validation (1,275 images), and 15% testing (1,275 images).
Results: The proposed ResNet50 model achieved 96.4% overall classification accuracy, with precision, recall, and F1-scores exceeding 94% across disease classes. Early-stage blast detection accuracy reached 92.3%, enabling disease identification 5–7 days before visible symptom manifestation. Area under receiver operating characteristic curve (ROC-AUC) was 0.981. Comparative analysis demonstrated superior performance compared with traditional machine learning (79.6%) and baseline CNN architectures (91.2%). Transfer learning reduced training time by 68% compared with training from scratch.
Conclusion: Deep learning-enabled rapid, non-invasive, and highly accurate detection of rice leaf blast, facilitating early disease intervention and reducing fungicide application by approximately 35% through improved disease surveillance. Integration with mobile applications and UAV platforms offers transformative potential for precision crop protection and sustainable rice production.