English

中国农机化学报

中国农机化学报 ›› 2023, Vol. 44 ›› Issue (8): 168-173.DOI: 10.13733/j.jcam.issn.2095-5553.2023.08.023

• 农业信息化工程 • 上一篇    下一篇

单目多视角图像的美国白蛾三维模型重建系统与试验

舒卓1, 2,陈立平1, 2,陈梅香2,张瑞瑞2,郭新宇3,温维亮3   

  1. 1. 西北农林科技大学机械与电子工程学院,陕西杨凌,712100;
    2. 国家农业智能装备工程技术研究中心,北京市,100097;
    3. 国家农业信息化工程技术研究中心/数字植物北京市重点实验室,北京市,100097
  • 出版日期:2023-08-15 发布日期:2023-09-12
  • 基金资助:
    国家自然科学基金面上项目(31971581);北京市农林科学院创新能力建设专项(KJCX20200206);北京市农林科学院杰出科学家培育专项(JKZX201903)

3D reconstruction system and precision evaluation of Hyphantria cunea (Drury) based on monocular and multiview images

Shu Zhuo1, 2, Chen Liping1, 2, Chen Meixiang2, Zhang Ruirui2, Guo Xinyu3, Wen Weiliang3   

  • Online:2023-08-15 Published:2023-09-12

摘要: 为构建具有高清纹理特征的美国白蛾三维模型,并对美国白蛾三维模型进行精度评价,开展单目多视角图像的美国白蛾三维重建的研究。首先构建昆虫多视角图像采集系统,实现对美国白蛾高质量序列图像自动化获取,建立美国白蛾高质量序列图像样本库。然后利用SFM与MVS相结合的方法,实现美国白蛾精准三维模型重建。最后以美国白蛾形态参数对重建的美国白蛾三维模型进行精度评估试验。结果表明,待评估值与参照值的相对误差小于5%、R2大于0.95。所重建的美国白蛾精准三维模型可以为深度学习样本扩增、植保病虫害识别与防控提供重要基础数据。

关键词: 三维重建;多视角图像;机器视觉;美国白蛾;重建效果评价, 虫害防控

Abstract: In order to construct the 3D model of H.cunea with highdefinition texture features, and to evaluate the accuracy of the 3D model of H.cunea objectively and quantitatively, the research and experiment of 3D reconstruction of H.cunea was carried out in this paper. Firstly, the multiview images acquisition system for insect was constructed. The multiview images acquisition system was used to acquire highquality sequence images of H.cunea, and the sample library of  highquality sequence image of H.cunea was established. Secondly, Structure from Motion (SFM) and Multiple View Stereo (MVS) algorithm were combined for the accurate 3D model reconstruction of H.cunea. Finally, the accuracy evaluation experiment of the reconstructed 3D model of H.cunea was carried out by morphological parameters. The experimental results showed that the relative error between the evaluated value and the reference value was less than 5%, and the R2 was greater than 0.95. The precise 3D model of H.cunea reconstructed in this paper can provide basic and intuitive data for deep learning sample augmentation, identification and control of plant diseases and pests.

Key words: 3D reconstruction, multiview images, machine vision, Hyphantria cunea (Drury), reconstruction effect evaluation, pest control

中图分类号: