Field-scale single-plant sunflower head detection and geometric parameters measurement integrating multi-modal UAV data, deep learning and point cloud analysis

Accurate monitoring of sunflower heads is critical for yield prediction, yet traditional methods are labor-intensive. This study proposed a novel framework integrating UAV remote sensing, deep learning, and point cloud analysis to address this challenge. The proposed method used a Dual-Branch YOLOv10n model, leveraging multi-modal data for precise detection of sunflower heads at various growth stages. Feature indices were designed and a two-step clustering technique was applied to extract sunflower head point clouds, from which geometric parameters such as diameter and volume are computed. The detection model achieved high accuracy (precision: 0.9, recall: 0.894, mAP@50: 0.932) across growth stages. A strong correlation (R 2 = 0.80) was found between diameter measurements from point cloud and ground-truth data, while volume showed good alignment with biomass (R 2 = 0.61). This method offers an innovative, efficient solution for field-scale crop monitoring and yield estimation, advancing agricultural practices. • Achieved sunflower head detection by integrating multi-modal data. • Designed feature indices of sunflower heads for different growth stages. • Designed a framework for accurate extraction of sunflower head point clouds. • Measured the geometric parameters of individual sunflower head at the field scale.