Publications

Deng, L; Chen, Y; Zhao, Y; Zhu, L; Gong, HL; Guo, LJ; Zou, HY (2021). An approach for reflectance anisotropy retrieval from UAV-based oblique photogrammetry hyperspectral imagery. INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION, 102, 102442.

Abstract
Reflectance Anisotropy (RA) contains crucial information about the optical behavior and structure of ground objects and is typically described by the Bidirectional Reflectance Distribution Function (BRDF). However, traditional RA-retrieval methods from satellites or the ground are constrained due to the low spatial resolution, illumination conditions, and instrument-associated limitations. In this study, Unmanned Aerial Vehicle (UAV) based oblique photography technology is applied to RA retrieval, aiming to explore the feasibility of UAV to obtain spatially continuous RA data with high spatial resolution and high accuracy. The Monte Carlo method is used to select and optimize the combination of multi-angle observation data of a variety of ground objects obtained by UAV-based oblique photogrammetry. The accuracy and applicability of two BRDF inversion models, namely the linear semi-empirical kernel-driven (LSEKD) model and the non-linear Rahman-Pinty-Verstraete (RPV) model, are analyzed, and the effects of different sampling window sizes on the BRDF results of various ground objects are compared. The main conclusions are: 1) Oblique photogrammetry is a highly efficient method for RA measurement at the centimeter-level resolution for spatially continuous regions, because it enables cameras with a narrow field of view (FOV) to obtain observation data with a wider viewing angle and more directions, ensuring the robustness and accuracy of the BRDF inversion model, with RMSE in the visible and near-infrared bands are about 0.003 and 0.019 (8-14%); 2) Both the LSEKD and the RPV model are suitable for the inversion of the BRDF with comparable accuracy, but the kernel functions for the LSEKD need to be carefully chosen in advance according to the characteristics of the ground object, while the RPV is adaptive for most objects; 3) The Monte Carlo method can ensure that the multi-angle observation data is distributed evenly in the hemisphere, thus providing the optimal dataset for a high-accuracy BRDF inversion; and 4) The RA of ground objects may change with the change of spatial resolution; the extremely high-resolution RA provides the ability to study ground objects at a finer scale. This study expands the methods for obtaining high-spatial resolution, high-accuracy, and spatially continuous RA. It has significant application potential in areas such as quantitative remote sensing, calibration of high-accurate remote-sensing products, and bridging the scale gap between satellite and ground-collected RA data.

DOI:
10.1016/j.jag.2021.102442

ISSN:
1569-8432