Scholarlink

Purpose: <p> Accurate estimation of crop plant height using unmanned aircraft system (UAS) imagery combined with structure-from-motion (SfM) photogrammetry is critical for plant phenotyping and precision agriculture; however, few studies have systematically evaluated how camera type, spectral configuration, and ground sampling distance (GSD) influence height estimation accuracy. This study assessed the performance of consumer-grade and scientific multispectral cameras for cotton plant height estimation under different flight altitudes, image overlaps, and crop growth stages.</p> Methods: <p> Two consumer-grade cameras (Nikon D7100 RGB and an NIR-modified Nikon D7100) and a scientific multispectral camera (MicaSense Altum) were evaluated over two growing seasons. A total of 105 datasets representing seven point cloud products and 15 flight configurations were generated from UAS imagery acquired at multiple flight altitudes. Plant height was estimated from the 99th percentile of point cloud elevations using ground surfaces derived from both GNSS measurements and image-based digital terrain models (DTMs). Estimation accuracy was compared across cameras, spectral bands, GSDs, and terrain modeling approaches.</p> Results: <p> The Nikon NIR camera consistently outperformed the Nikon RGB camera, reducing the mean RMSE from 12.9 cm to 7.0 cm. For the Altum camera, the Red edge and NIR bands produced substantially lower RMSEs (9.7 and 9.0 cm, respectively) than the visible bands (15.9–25.8 cm). Merging Nikon RGB and NIR point clouds did not improve accuracy, whereas combining the Altum Red edge and NIR bands reduced the RMSE to 8.6 cm, with no additional improvement from using all five bands. Nikon NIR provided stable plant height estimates across all evaluated flight altitudes (30–120 m; 0.5–2.0 cm GSD), whereas Altum achieved its best performance at lower altitudes (30–60 m; 1.3–2.6 cm GSD). Under these conditions, the Altum NIR band produced the highest accuracy (RMSE = 5.6 cm), outperforming the Nikon NIR camera (RMSE = 7.3 cm). Compared with point cloud classification, DSM-filtered DTMs produced smaller and more consistent deviations from GNSS-derived ground elevations while providing comparable plant height estimates.</p> Conclusions: <p>Camera spectral sensitivity, spatial resolution, and terrain modeling methods significantly influence UAS-based cotton plant height estimation. The Nikon NIR camera offers robust and cost-effective performance across a wide range of flight altitudes, whereas the MicaSense Altum provides superior accuracy when flown at finer GSDs. DSM-filtered DTMs represent a reliable alternative to GNSS-derived ground elevations for plant height estimation. These findings provide practical guidance for selecting imaging sensors, optimizing flight parameters, and choosing terrain modeling approaches for UAS-based crop phenotyping.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Evaluating three cameras for cotton plant height estimation using UAS-derived point clouds

  • 期刊论文

  • 发表日期：

    2026年7月22日

查看全文
  • Chenghai Yang,
  • Charles P.-C. Suh,
  • Bradley K. Fritz

摘要

Purpose:

Accurate estimation of crop plant height using unmanned aircraft system (UAS) imagery combined with structure-from-motion (SfM) photogrammetry is critical for plant phenotyping and precision agriculture; however, few studies have systematically evaluated how camera type, spectral configuration, and ground sampling distance (GSD) influence height estimation accuracy. This study assessed the performance of consumer-grade and scientific multispectral cameras for cotton plant height estimation under different flight altitudes, image overlaps, and crop growth stages.

Methods:

Two consumer-grade cameras (Nikon D7100 RGB and an NIR-modified Nikon D7100) and a scientific multispectral camera (MicaSense Altum) were evaluated over two growing seasons. A total of 105 datasets representing seven point cloud products and 15 flight configurations were generated from UAS imagery acquired at multiple flight altitudes. Plant height was estimated from the 99th percentile of point cloud elevations using ground surfaces derived from both GNSS measurements and image-based digital terrain models (DTMs). Estimation accuracy was compared across cameras, spectral bands, GSDs, and terrain modeling approaches.

Results:

The Nikon NIR camera consistently outperformed the Nikon RGB camera, reducing the mean RMSE from 12.9 cm to 7.0 cm. For the Altum camera, the Red edge and NIR bands produced substantially lower RMSEs (9.7 and 9.0 cm, respectively) than the visible bands (15.9–25.8 cm). Merging Nikon RGB and NIR point clouds did not improve accuracy, whereas combining the Altum Red edge and NIR bands reduced the RMSE to 8.6 cm, with no additional improvement from using all five bands. Nikon NIR provided stable plant height estimates across all evaluated flight altitudes (30–120 m; 0.5–2.0 cm GSD), whereas Altum achieved its best performance at lower altitudes (30–60 m; 1.3–2.6 cm GSD). Under these conditions, the Altum NIR band produced the highest accuracy (RMSE = 5.6 cm), outperforming the Nikon NIR camera (RMSE = 7.3 cm). Compared with point cloud classification, DSM-filtered DTMs produced smaller and more consistent deviations from GNSS-derived ground elevations while providing comparable plant height estimates.

Conclusions:

Camera spectral sensitivity, spatial resolution, and terrain modeling methods significantly influence UAS-based cotton plant height estimation. The Nikon NIR camera offers robust and cost-effective performance across a wide range of flight altitudes, whereas the MicaSense Altum provides superior accuracy when flown at finer GSDs. DSM-filtered DTMs represent a reliable alternative to GNSS-derived ground elevations for plant height estimation. These findings provide practical guidance for selecting imaging sensors, optimizing flight parameters, and choosing terrain modeling approaches for UAS-based crop phenotyping.