همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

Comparative Evaluation of YOLO-Based and CNN Models for Wheat-Head Detection in Precision Agriculture

نویسندگان :

Shahin Rostami¹ Shervin Danesh² Niloofar Mirzaei Chahardeh³

1- دانشگاه آزاد اسلامی واحد کرج 2- دانشگاه آزاد اسلامی واحد کرج 3- دانشگاه آزاد واحد علوم تحقیقات تهران

کلمات کلیدی :

wheat-head detection،deep learning،precision agriculture،YOLOv11،real-time object detection

چکیده :

Accurate wheat-head detection is essential for yield estimation, crop monitoring, and automated decision-support systems in precision agriculture. This study presents a comprehensive comparative evaluation of four deep learning models YOLOv8, YOLOv11, Faster R-CNN and MobileNetV2 for wheat-head detection using the Global Wheat 2020 dataset under unified training and testing conditions. Our analysis focuses on detection accuracy, inference speed, and deploy ability on edge platforms for real-time field applications. The models are evaluated using precision, recall, F1-score, mAP@0.5 and inference time under a unified training pipeline. Among the evaluated models, YOLOv8 demonstrated the best overall balance between accuracy and inference speed, achieving a state-of-the-art mAP@0.5 of 0.96 while still remaining efficient for real-time deployment. YOLOv11, implemented as the latest Ultralytics one-stage detector, provides an effective balance between accuracy and computational efficiency for wheat-head detection. Faster R-CNN offered higher precision but suffered from slow inference, which hinders real-time deployment, whereas MobileNetV2 provided faster computation but lower precision. A light ablation analysis indicates that transformer-based feature fusion and dynamic convolution further improve YOLOv11’s robustness in dense field scenes. Unlike prior work, this study offers a unified, side-by-side comparison of YOLOv8, YOLOv11, Faster R-CNN and MobileNetV2 on the GlobalWheat2020 dataset, establishing a new state-of-the-art mAP@0.5 of 0.96 for wheat-head detection.