Crop Height Estimation of Sorghum From High Resolution Multispectral Images Using the Structure From Motion (SFM) Algorithm

Abstract

Crop height (CH) is the key indicators of crop growth, biomass and yield. However, obtaining CH information with manual measurement is inefficient for larger areas. High-resolution unmanned air vehicle (UAV) images offer a new alternative to traditional CH measurements. In this study, we compared three approaches to estimate sorghum CH using high-resolution multispectral images based on structure from motion (SfM) algorithm and spectral vegetation indices. In the first approach, CH was estimated based on the difference between the Digital Surface Model (DSM) map and Digital Terrain Model (DTM) map generated from UAV images captured immediately after the sowing. In the second approach, DTM was generated from DSM. In the last approach, CH was estimated using the spectral vegetation indices. High-resolution multispectral images were obtained at 40 m above ground level elevation. Ground control points were laid around the study area, and these point positions were determined using a GPS device. DSM and DTM images were generated from 3D point cloud data and the SfM algorithm. Results showed that the SfM technique could estimate sorghum CH accurately using DSM, DTM and GCPs (R2 = 0.97, RMSE = 8.77 cm, MAPE = 5.98%). Also, a high correlation was observed between estimated and measured sorghum CH using DTM maps generated from DSM maps (R2, RMSE, MAPE were 0.94, 12.2 cm, 6.66%). Moreover, GNDVI was the best vegetation index to estimate sorghum CH (R2 = 0.81, RMSE = 24.6 cm, MAPE = 12.56%). Overall, this study demonstrates the UAV potential for CH estimates and reducing the cost of obtaining CH information.