沈阳农业大学学报2025，Vol.56Issue(6)：20-30,11.DOI:10.3969/j.issn.1000-1700.2025.06.003
一种基于无人机RGB图像的玉米雄穗检测方法
A Detecting Method for Maize Tassels Based on UAV RGB Images
摘要
Abstract
[Objective]Accurate identification and detection of maize tassels are crucial for improving detasseling efficiency.Addressing the current limitations of insufficient detection accuracy and poor robustness of deep learning algorithms in complex field environments,this paper proposes an enhanced YOLOv8-based method for efficient maize tassel detection.[Methods]A detection dataset with strong generalization capabilities was constructed by collecting multi-scenario data via UAVs under varying weather conditions and flight altitudes.By incorporating a"Ghost Convolution"(GhostConv)module into the backbone network of YOLOv8,and adding an Attentional Scale Sequence Fusion(ASF)module and a Gather-and-Distribute Mechanism(Gold)module to the neck network,the model can better extract target features and enhances performance in object localization and regression tasks.The improved detection model is named as AG-YOLO.[Results]AG-YOLO demonstrates outstanding performance in maize tassel detection,achieving an Average Precision(mAP)of 89.3%while reducing model size by approximately 18.3%compared to the original model.This performance significantly outperforms other mainstream detection algorithms such as YOLOv3,YOLOv5,YOLOv6,and YOLOv8.Particularly in scenarios such as early tasseling stages,heavy leaf occlusion,dense target distribution,or complex backgrounds,AG-YOLO demonstrated outstanding detection capabilities.[Conclusion]The improved AG-YOLO model effectively detects maize tassels in complex and variable field environments,balancing high detection accuracy,lightweight model size,and strong robustness.It exhibits significant practical value in practical applications.This study provides efficient and reliable technical support for automated and intelligent maize detasseling operations,laying a solid foundation for optimizing and promoting precise phenotyping detection models for other crops in future smart agriculture applications.关键词
深度学习/YOLOv8/玉米雄穗/图像识别Key words
deep learning/YOLOv8/maize tassel/image recognition分类
农业科技引用本文复制引用
曹丽英,钟舸澳,赵浩宇,毕红杰..一种基于无人机RGB图像的玉米雄穗检测方法[J].沈阳农业大学学报,2025,56(6):20-30,11.基金项目
吉林省科技发展计划项目中青年科技创新人才(团队)培育项目(20250601061RC) （团队）
