Influence of ground control points and processing parameters on UAS image mosaicking for plant height estimation

Authors: Yang, Chenghai; ZHAO, HENGQIAN - China University Of Mining And Technology; GUO, WEI - Henan Agricultural University; ZHANG, JIAN - Huazhong Agricultural University; Suh, Charles; Fritz, Bradley

Submitted to: International Conference on Precision Agriculture Abstracts & Proceedings
Publication Type: Proceedings
Publication Acceptance Date: 6/28/2024
Publication Date: 7/21/2024
Citation: Yang, C., Zhao, H., Guo, W., Zhang, J., Suh, C.P., Fritz, B.K. 2024. Influence of ground control points and processing parameters on UAS image mosaicking for plant height estimation. International Conference on Precision Agriculture, 21-24 July 2024, Manhattan, KS. pp, 1-13.

Interpretive Summary: Three-dimensional (3D) point clouds and digital surface models (DSMs), generated using images from unmanned aircraft systems (UAS), are often used for plant height estimation in phenotyping and precision agriculture. This study examined how the quantity and placement of ground control points (GCPs) and image processing parameters affected the accuracy of 3D point clouds and DSMs for estimating plant height using UAS images from a 2-hectare experimental field with four crops. The images were processed with various GCP configurations and processing parameters to generate image products, including 3D point clouds and DSMs. Results showed that increasing GCPs significantly improved positional accuracy up to a certain point, but adding more than five GCPs offered minimal additional benefit. Plant height estimates derived from 3D point clouds were found to be more precise and reliable than those from DSMs. These results provide valuable guidelines for optimizing GCP use and image processing settings, enhancing plant height measurement accuracy in agricultural research and precision farming.

Technical Abstract: Three-dimensional (3D) point clouds and digital surface models (DSMs), generated using overlapping images from unmanned aircraft systems (UASs), are often used for plant height estimation in phenotyping and precision agriculture. This study examined the effects of the quantity and placement of ground control points (GCPs) and image processing parameters on the creation of 3D point clouds and DSMs for plant height estimation. A 2-ha field containing multiple experimental plots with four crops (corn, cotton, sorghum, and soybean) was used for this study. Thirty-six panels were systematically positioned across these plots, with 12 at ground level and 12 each at approximately 0.75 m and 1.5 m above ground. Aerial images were captured at 60 m above ground level using a rotary hexacopter equipped with a Nikon D7100 camera. Plant height was manually measured from 48 sampling points among the four crops. Orthomosaics, 3D point clouds, and DSMs with various GCP configurations and processing parameter combinations were generated using Pix4Dmapper software. Results showed that increasing GCPs from one to five reduced the total positional root mean square error (RMSE) from 2.3 m to 3 cm. However, adding more GCPs only marginally reduced RMSE to approximately 2-3 cm. Panel positions on or above the ground did not notably affect positional accuracy. Statistical analysis showed that the correlation coefficients between ground-measured and point cloud-extracted values were similar and consistent with four or more GCPs. Moreover, point cloud-based plant height estimates were more accurate and consistent than DSM-based estimates. Changing image processing settings, including keypoint image scale, densification image scale, point density, and minimum number of point-to-image matches, directly affected the number of 3D points created, processing time, and plant height estimation. The results from this study provide practical guidance for selecting suitable GCPs and image processing parameters for accurate plant height estimation with UAS imagery in test plots and small fields.