Precision detection and geolocation of missed pre-tassels in hybrid maize seed production using UAV-based deep learning

Abstract

Hybrid maize seed production relies on detasseling, a critical process to ensure genetic purity by removing male pre-tassels from female plants. However, missed pre-tassels, which are immature tassels partially enclosed by leaves and similar in color to maize foliage, remain difficult to detect and typically require labor-intensive manual inspection. This study proposes an improved UAV-based detection framework, YOLO for Missed Pre-Tassel (YOLO-MPT), built upon YOLOv7 for precise identification and geolocation of missed pre-tassels in hybrid maize fields. YOLO-MPT integrates deformable convolutions (DCNv2) for adaptive feature extraction, the S²-MLPv2 attention mechanism for enhanced spatial representation, and an additional small-object detection head to increase sensitivity to tiny or occluded targets. A comprehensive UAV-derived pre-tassel dataset was constructed under diverse agronomic and lighting conditions to support model training and validation. The impact of input image size on detection performance was systematically analyzed to identify the optimal training resolution. Experimental results show that YOLO-MPT achieved an average precision (AP) of 93.8 %, precision (P) of 93.3 %, recall (R) of 90.2 %, and an F1-score of 91.7 %, outperforming baseline models. Furthermore, a geographic coordinate extraction method was developed and integrated into a standalone “Missed Pre-Tassel Detection and Localization Software,” enabling automatic conversion of pixel detections into precise geospatial locations. Field experiments verified the workflow’s robustness and positioning accuracy, demonstrating the system’s potential to improve post-detasseling efficiency and quality assurance in hybrid maize seed production.

Plant phenotyping relevance

UAV画像からトウモロコシの未抽苔を検出・地理定位する手法を開発し、データセット、性能検証、ソフトウェア化まで行っており、植物状態の取得・抽出が研究の中心である。

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.