YOLO-detassel: Efficient object detection for Omitted Pre-Tassel in detasseling operation for maize seed production

文献类型： 外文期刊

第一作者： Yang, Jiaxuan

作者： Yang, Jiaxuan;Zhang, Ruirui;Ding, Chenchen;Chen, Liping;Xie, Yuxin;Ou, Hong;Yang, Jiaxuan;Zhang, Ruirui;Ding, Chenchen;Chen, Liping;Xie, Yuxin;Ou, Hong;Yang, Jiaxuan;Chen, Liping

作者机构：

关键词： Detasseling; Object detection; UAV; Deep learning; Maize hybrid seed production

期刊名称：COMPUTERS AND ELECTRONICS IN AGRICULTURE （ 影响因子：8.9； 五年影响因子：9.3 ）

ISSN： 0168-1699

年卷期： 2025 年 231 卷

页码：

收录情况： SCI

摘要： Maize (Zea mays L.) stands as a pivotal grain crop, and hybrid seed production enables the combination of desirable traits from diverse parent varieties. Ensuring seed purity is fundamental in hybrid seed production, where accurate unreleased tassel detection and efficient detasseling are crucial. Specifically, in the context of maize field seed production, the detasseling of the female parent necessitates prior execution preceding pollen shedding. Despite concerted efforts encompassing both mechanical and manual detasseling methodologies, an estimated range of 420-480 tassels per acre evade removal, exerting a detrimental impact on the purity of hybrid seeds. While extant researches concerning maize tassel detection and enumeration primarily concentrate on postpollen shedding stages, leaving pre-tassel identification largely unexplored. To address this gap, this study develops a novel object detection framework, YOLO-detassel, for pre-tassel identification during the tasseling stage. UAVs were deployed to capture high-resolution RGB imagery from a hybrid seed production field in western China, creating a dataset of pre-tassel maize plants. YOLO-detassel enhances the YOLOv5 model through several innovations: (1) MobileNetV3 is integrated into the backbone to reduce parameters and improve computational efficiency while maintaining high detection accuracy; (2) the Simple Attention Module (SimAM) incorporates three-dimensional weights to better capture spatial and channel features holistically; and (3) the Content-Aware Reassembly of Features module boosts context integration and spatial awareness for robust detection in complex field conditions. Empirical evaluations attest to the efficacy of the proposed YOLO-detassel algorithm, attaining an Average Precision (AP) of 96.8% for pre-tassel object detection, coupled with a commendable recall rate of 94.2%, precision of 98.4% and a notable detection speed of 32 frames per second. The elucidated object detection algorithm not only signifies an innovative stride towards addressing the challenge of identifying missed pre-tassels in maize seed production but also holds promise for extending its applicability towards the detection of pre-tassels on female parent maize plants leveraging UAV imagery automatically.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序