Crop diseases remain a critical threat to global food security, contributing to substantial yield losses and reduced farmer incomes. Timely and accurate identification of these diseases is essential to mitigate their impact. Traditional diagnostic methods, dependent on expert visual inspection, are labour-intensive, time-consuming, and prone to judgment errors. Accurate and timely detection of crop diseases supports sustainable agricultural management and contributes to achieving global objectives under the United Nations Sustainable Development Goal 2 on Zero Hunger. This study proposes a three step framework that relies on pattern recognition and classification of visual disease symptoms to deliver reliable, field-applicable diagnostics. The approach combines image acquisition through smartphone camera with a structured processing pipeline that includes feature extraction, classification, and result delivery via a mobile application built on a three-tier architecture. Convolutional Neural Networks and an optimized VGG-16 model form the core classification engine, trained to recognize 19 leaf based diseases across wheat, rice, fodder, maize, and sugarcane. The models were trained and evaluated on a dataset comprising both field-collected and publicly available images using repeated stratified k-fold cross-validation. The framework achieves accuracies of 84.61% for wheat, 44.15% for rice, 85.71% for fodder, 95.23% for maize, and 64.28% for sugarcane (testing accuracy of the best-performing model per crop, where VGG-16 demonstrated superior generalization). The framework is able to support farmers, by integrating a technically robust backend with a simple and oriented interface, with diagnosis of multiple crops from a single platform, offering a scalable solution for precision agriculture and sustainable crop protection.
Kochhar et al. (Mon,) studied this question.