ZamYOLO-Maize: A YOLOv8n-Based Deep Learning Framework for Automated Detection and Classification of Maize Leaf Diseases in Field Conditions in Zambia

Prudence Kalunga^1*Douglas Kunda²Aaron Zimba¹

• ¹ZCAS University, Zambia, Lusaka, Zambia
• ²DMI-St Eugene University, Lusaka, Zambia

Maize, a critical staple crop in Zambia, faces persistent threats from foliar diseases such as Gray Leaf Spot, Northern Corn Leaf Blight, and Maize Streak Virus. Smallholder farmers, constrained by complex field conditions and limited access to expert diagnostics, require accessible, real-time solutions for accurate disease identification. This study addresses this gap through a dual contribution: first, the creation of a novel, field-captured dataset of Zambian maize leaf images, annotated with bounding boxes for disease lesions and classified by disease type and severity, a resource tailored to local agri-ecological conditions. Second, we introduce the ZamYOLO-Maize framework, an automated diagnostic system built upon this dataset. The framework integrates a comparative analysis of four cutting-edge deep learning models being YOLOv5n, YOLOv8s, YOLOv10s, and YOLOv8n for lesion detection, followed by hierarchical classification and severity assessment modules. Our experiments demonstrate that YOLOv8n offers the optimal balance for edge deployment, with the fastest inference speed (4.65 ms/img) and a competitive F1-score (0.995), while YOLOv10s achieves the highest overall predictive performance (Precision = 0.997, Recall = 0.999, F1-score = 0.999). The system's robustness in handling field variability including occlusions, lighting changes, and diverse symptom presentations validates its practical utility. By unifying a purpose-built dataset with an efficient, multi-stage deep learning pipeline, this work establishes a scalable foundation for mobile-based diagnostic tools, promising to enhance precision agriculture and support food security initiatives across Zambia and similar agricultural regions.