COMPARATIVE BENCHMARKING OF YOLOV11 AND MACHINE LEARNING MODELS FOR MAIZE LEAF DISEASE DETECTION

Abstract

Maize diseases are a major threat to food security, particularly in developing regions where early detection is critical to minimizing yield losses. Traditional machine-learning models such as Convolutional Neural Networks (CNN), K-Nearest Neighbors (KNN), Random Forests (RF), and Support Vector Machines (SVM) have been widely applied but face limitations in scalability, robustness, and real-time deployment. Enhanced models, including EfficientNet, XGBoost, and LS-SVM, have improved accuracy yet remain computationally demanding. To address these gaps, this study evaluates the latest YOLOv11 object detection framework for maize disease classification. A dataset of 750 annotated maize leaf images spanning four classes (Blight, Rust, Gray Leaf Spot, and Healthy) was used for benchmarking against both classical and enhanced baselines. Results show that YOLOv11 significantly outperformed all other models, achieving 99.8% accuracy, with near-perfect precision, recall, and F1-scores, alongside the fastest inference time of 12 ms per image. These findings highlight YOLOv11’s capability to combine accuracy and efficiency, making it suitable for real-time deployment on mobile devices and drone-based platforms. The study makes three key contributions: (i) establishing a rigorous benchmarking framework that fairly compares classical, enhanced, and state-of-the-art models; (ii) demonstrating YOLOv11’s superior performance in both accuracy and inference speed; and (iii) underscoring its potential applications in precision agriculture. This research provides compelling evidence that YOLOv11 represents a transformative advancement in crop disease detection. Its integration into mobile advisory systems, drone-based surveillance, and decision-support tools can directly contribute to sustainable agriculture and global food security.