AI-Based System for Automatic Detection of Gray Leaf Spot Disease in Maize Crops

Gray Leaf Spot (GLS) disease, caused by the fungal pathogen Cercospora zeae-maydis, is a devastating foliar disease that threatens maize production in the southern highlands of Tanzania, particularly in the Iringa region, where more than 500,000 smallholder farmers depend on maize for food security. Existing diagnostic practice relies on visual inspection by agricultural extension officers, each of whom serves an estimated 5,000 farmers, which delays diagnosis and allows yield losses to accumulate. This paper presents MaizeGuard AI, an AI-based system for automatic GLS detection built around a Convolutional Neural Network (CNN) approach. The system uses a dual-engine architecture: a primary MobileNetV2 CNN that delivers high-accuracy inference when connectivity is available, and a lightweight Random Trees (RTrees) ensemble that performs offline detection in the field. A dataset of 1,953 maize leaf images – 1,466 images drawn from a public Kaggle repository and 487 images collected in Iringa – was used to train and evaluate both engines through two-phase transfer learning. On the binary healthy-versus-diseased task the MobileNetV2 model achieved 92.62% accuracy, 98.90% precision, 81.08% recall and an 89.11% F1-score, while GLS-specific detection reached 96.55% accuracy. The system is deployed as a bilingual (English/Swahili) Laravel web application that overlays Grad-CAM heatmaps and quantitative donut-chart severity indicators on every uploaded leaf image, giving farmers and extension officers an explainable result within seconds. The findings show that CNN-based automated disease detection is both technically feasible and practically useful for strengthening food security and shortening diagnostic delay in resource-constrained Tanzanian agricultural settings.