Machine learning model for sugarcane disease detection and classification to improve yield

Abstract

The imperative to sustain agricultural productivity amid escalating threats from plant diseases and nutrient deficiencies is critical for global food security, particularly in sugarcane-dependent regions where manual diagnostic methods are error-prone and inaccessible. This study bridges this gap by developing a deep learning model that automates the detection of six critical sugarcane stressors red rot, using computer vision. Leveraging a dataset of 12,300 images (balanced across healthy and diseased leaves), a convolutional neural network (CNN) was trained with hyperparameter optimization (Adam optimizer, learning rate = 0.001) and data augmentation techniques (rotation, flipping, normalization) to enhance robustness against field variability. The model achieved 85% overall accuracy on a holdout test set, with class-specific precision scores of 93.3% for red rot (distinct necrotic lesions), 93.8% for smut (whip-like fungal structures), and 85.7% for leaf scald, while nitrogen deficiency detection lagged at 16.7% accuracy due to symptom overlap with early-stage leaf scald chlorosis. Performance metrics, including 89.5% precision, 91% recall, and a weighted F1-score of 90.2%, were validated through a confusion matrix, revealing misclassification patterns and underscoring the need for complementary soil health data. Deployed on cloud-based infrastructure (Google Colab), the system processes images in 3 milliseconds per prediction, offering farmers in resource-limited regions a scalable, real-time diagnostic tool to mitigate yield losses by 20 to 30%. Challenges such as class imbalance (e.g., mosaic virus accuracy inflation from limited samples) and environmental variability were addressed through methodological rigor, including stratified sampling and augmented training. Future directions include expanding the model to analyze stems and roots, integrating IoT soil sensors, and optimizing edge deployment for offline use. By aligning with Sustainable Development Goal 2 (Zero Hunger), this research demonstrates how AI-driven innovations can democratize access to precision agriculture, empowering smallholder farmers to adopt proactive disease management strategies and enhance crop resilience in the face of climate and pathogen pressures.