Comparison of Crop Surface Models and 3D Point Clouds by UAV Imagery on Estimating Plant Height and Biomass Volume of Pasture Grass

Abstract

Plant biomass is considered an important parameter for crop management and yield estimation, especially for grassland. Aerial photogrammetric techniques have been used for vegetation data gathering of the areas of dense vegetation fields with high research interest. Recent advances in computer vision include structure from motion and multi-view stereopsis (SfM-MVS) techniques, which can derive 3D data such as digital surface models (DSMs) and orthomosaic from overlapping photography taken from multiple angles. The difference between the DSMs of a planted field and the digital terrain model (DTM) has been referred to crop surface model (CSM). Ever since SfM-MVS has been adopted to derive plant height (PH) and above-ground biomass using CSMs at 2013, this method has become the most explored and verified approach to simulate the structure of crops all over the world. However, the complexity of crop structure is thought to be not well represented in DSMs because the DSMs have only one Z value at each 2D pixel. Besides, lacking a DTM representing the bare ground is another problem when adopting the CSM method. On the other hand, the 3D point cloud where DSMs are derived from UAV may provide the structure information in a faster and more detailed way. This research tested the capability of 3D point cloud in estimating plant height and biomass volume of pasture grass, and compared the results with CSMs. UAV photography were conducted at the experimental field of Obihiro University of Agriculture and veterinary Medicine, Hokkaido, Japan, 2019. The biomass volume estimated by DSM and point cloud have no significant difference, showing that DSM and point cloud have the same performance at estimating biomass volume of grass. In the case that only the simple value data is required, the point cloud data is recommended.