地栽草莓机器人检测预测去除式气吹去遮挡技术
CSTR:
作者:
作者单位:

作者简介:

通讯作者:

中图分类号:

基金项目:

国家重点研发计划项目(2025YFE0209300)、浙江省自然科学基金重大项目(LD24E050006)和国家自然科学基金项目(32372004)


Detection of Information Obstruction in Field-planted Strawberries and Prediction of Parameters for Removing Obstruction by Air Blowing
Author:
Affiliation:

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    针对地栽草莓种植中枝叶遮挡导致目标检测失效的问题,本文提出融合实例分割与遮挡检测,遮挡预测与去遮挡气流优化的协同算法。首先,构建基于YOLO 11 seg 的草莓果实/ 遮挡物实时分割模型,生成复杂场景下草莓果实、遮挡物掩膜完整提取,随后分析检测当前草莓遮挡率;在此基础上,对于非重度遮挡目标区域(当前遮挡率小于等于70% ),系统直接以草莓果实掩膜的几何中心为气吹靶点,无需启动螺旋搜索区域生长算法。针对重度遮挡目标区域(当前遮挡率大于70% ),开发螺旋搜索区域生长模型搜索算法,定位最优气吹干预区域,精准捕捉遮挡率演变的时间特征;并用轻量级CNN 以螺旋特征为输入,精确预测气吹后的遮挡率;最后,集成多参数可调气吹装置物理去遮挡物。在遮挡率预测方面,遮挡信息预测精度高(R2达0.925,RMSE为2.57% ),显著提升遮挡率预估精度和复杂环境适用性,通过整体方法的实施,包括检测、预测和气吹去除等多个步骤,在实地进行田间试验,试验表明,该方法在茎叶遮挡场景下能有效降低草莓果实遮挡率,验证了算法的有效性,为设施农业作物去遮挡提供“检测预测去除”解决方案。田间试验结果表明,该系统在90个严重遮挡样本中将平均遮挡率从68.5%降至12.8% ,82个样本 (91.1%)达到遮挡率小于15% ,显著提升了地栽草莓在复杂环境下的识别鲁棒性与机器人采摘适应性。

    Abstract:

    Stem and leaf obstruction, interlacing, and overlapping are common phenomena during the growth process of field-grown strawberries, posing significant challenges for fruit target detection by harvesting robots. To address target detection failures caused by foliage occlusion in strawberry cultivation, a synergistic algorithm integrating instance segmentation with occlusion detection, occlusion prediction, and airflow-optimized deoccultation was proposed. Firstly, a real-time segmentation model based on YOLO 11 seg was constructed to generate complete masks for strawberry fruits and occluding objects in complex scenes, followed by analysis to determine the current strawberry occlusion rate. For non-severely occluded regions (current occlusion rate no more than 70% ), the system directly targeted the geometric center of the strawberry mask for air-blowing intervention without initiating the spiral search-region growing algorithm. For heavily obstructed target areas (current obstruction rate greater than 70% ), a spiral search region growing model search algorithm was developed to locate the optimal airblowing intervention zone, precisely capturing the temporal characteristics of obstruction rate evolution. A lightweight CNN then used spiral features as input to accurately predict the post-air-blowing obstruction rate. Finally, a multi-parameter adjustable air-blowing device physically removed obstructions through integrated operation. Regarding occlusion rate prediction, the method achieved high accuracy in occlusion information forecasting (R2 was 0.925, RMSE was 2.57%), significantly enhancing estimation accuracy and adaptability to complex environments. Through the implementation of the integrated approach, i. e. , encompassing detection, prediction, and air-blowing removal, the method underwent field trials. Results demonstrated its effectiveness in reducing strawberry fruit occlusion rates under stemand-leaf shading scenarios, validating the algorithm's efficacy and providing a “ detection-predictionremoval” solution for crop de-occlusion in protected agriculture. Field trial results indicated that the system reduced the average occlusion rate from 68.5% to 12.8% across 90 severely occluded samples. In 82 samples (91.1% ), the occlusion rate fell below 15% , significantly enhancing the robustness of ground-grown strawberry identification and robotic harvesting adaptability in complex environments.

    参考文献
    相似文献
    引证文献
引用本文

马锃宏,董乃深,林熙淼,赵胤,顾峻瑜,杜小强,武传宇.地栽草莓机器人检测预测去除式气吹去遮挡技术[J].农业机械学报,2026,57(5):138-148. MA Zenghong, DONG Naishen, LIN Ximiao, ZHAO Yin, GU Junyu, DU Xiaoqiang, WU Chuanyu. Detection of Information Obstruction in Field-planted Strawberries and Prediction of Parameters for Removing Obstruction by Air Blowing[J]. Transactions of the Chinese Society for Agricultural Machinery,2026,57(5):138-148.

复制
分享
相关视频

文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2025-09-30
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2026-03-01
  • 出版日期:
文章二维码