Maize height estimation using combined unmanned aerial vehicle oblique photography and LIDAR canopy dynamic characteristics

Plant height (PH) is a key indicator for assessing plant health and growth, as well as an important criterion for ideotype breeding, making efficient, accurate PH measurement essential. The traditional manual measurement method is small-scale, inefficient, and time-consuming, which has restricted crop breeding progress and agricultural production efficiency improvements. Therefore, in this study, we aimed to estimate the maize populations' PH in the field using multisource unmanned aerial vehicle (UAV) images and to explore the factors that affect the accuracy of oblique photography (OP) and light detection and ranging (LIDAR) measurements of the populations' PH. To this end, we acquired red–green–blue (RGB), LIDAR, and multispectral image data for a maize population canopy using a UAV in four growth stages in two maize production regions, and we obtained the maize PH via field measurement as validation data. First, we reconstructed the three-dimensional point cloud using OP and LIDAR, and we used the 2nd and 100th percentiles of the elevation information as the lower and upper boundaries, respectively, to estimate the maize height. In addition, we used multispectral-based vegetation indices (VIs) and RGB-based texture indices (TIs) to describe the population canopy's characteristics, and we compared and analyzed the accuracy of the PH estimation for four data types and ten data fusion methods. The results revealed that all four data types could accurately estimate the maize PH and that the R2 values were all greater than 0.75. The light detection and ranging elevation (LIDAR_elev) had the highest estimation accuracy, and the R2 values of the maximum, minimum, and average PH of the population were all greater than 0.90. We also found that the dynamic characteristics of the canopy growth were important factors affecting the estimation of the PH using the oblique photography elevation (OP_elev) and LIDAR_elev. Thus, after fusing the VIs and TIs, the R2 value reached a maximum of 0.98, and the root mean square error (RMSE) was 8 cm. The use of different regression models had little influence on the estimation of the PH. Our research provides an effective method for high-throughput evaluation of a maize population's PH growth and serves as a reference for studying other phenotype parameters.